Rapid quotation generation method and system for virtual power plant participating in market under cooperation of main power plant, distribution unit and micro power plant

Abstract

The invention discloses a rapid quotation generation method and system for a virtual power plant participating in a market under main, distribution and micro cooperation. The method comprises the following steps: establishing a VPP operation cost minimum optimization model; solving a VPP feasible region by using a vertex search method to form training data; training a KAN network to realize batch prediction of a VPP feasible region; submitting the predicted VPP feasible region to the electricity market for optimal clearing settlement; and notifying each virtual power plant of the clearing result, and executing power generation by the virtual power plant according to the plan. According to the method, the operation feasible region of the virtual power plant is solved through the equivalent projection method, and integrated optimization of operation states and resources of different power grid levels is realized. And converting the target function and the constraint condition of the virtual power plant into the constraint condition of the coordination variable, and submitting the constraint condition to the power market for optimization and clearing. The conversion not only avoids the problem of slow convergence of a traditional iterative calculation method, but also promotes information sharing and decision coordination among all levels of power grids under main, distribution and micro coordination, and improves the operation efficiency of the whole power system.

Description

Technical Field

[0001] The present invention relates to the technical field of power systems, and particularly to a method and system for quickly generating bids of a virtual power plant participating in the power market under the cooperation of main grid, distribution grid and microgrid. Background Art

[0002] The bids of traditional virtual power plants participating in the power market include power generation and price information. Market operators conduct market clearing based on the bids and demands of all participants. This mechanism pays more attention to the interests of individual participants rather than the optimal operation of the entire system. To achieve the lowest comprehensive cost of the system, centralized optimization dispatching and distributed optimization clearing methods are usually adopted. Centralized optimization dispatching considering the cooperation of main grid, distribution grid and microgrid requires the unified dispatching center of the power grid to collect the parameter information of all devices and conduct optimization dispatching, but it involves privacy protection issues. Distributed optimization can effectively protect user privacy and relieve the computing and communication pressure of the unified dispatching center, but it requires multiple iterative solutions and the calculation convergence is slow. In addition, for the clearing optimization problem of large-scale power markets, due to the existence of a large number of 0-1 variables in the main grid, the existing commercial solvers have a slow solving speed.

[0003] For the day-ahead market (DAMs), it is crucial to propose solutions within a short time. Therefore, designing a new bidding method for VPP (i.e., virtual power plant) participating in the power market under the cooperation of main grid, distribution grid and microgrid to accelerate the optimization clearing of the power market has become an urgent problem to be solved. Summary of the Invention

[0004] In view of the above problems, the present invention provides a method for quickly generating bids of a virtual power plant participating in the power market under the cooperation of main grid, distribution grid and microgrid, which can effectively accelerate the bidding and clearing process of the virtual power plant under the cooperation of main grid, distribution grid and microgrid, and improve the flexibility and economy of the system.

[0005] The technical solution of the present invention is as follows: A method for quickly generating bids of a virtual power plant participating in the market under the cooperation of main grid, distribution grid and microgrid includes the following steps:

[0006] S1. Establish an optimization model for minimizing the operating cost of VPP;

[0007] S2. On the basis of step S1, use the vertex search method to solve the feasible region of VPP and form training data;

[0008] S3. Train the KAN network to realize batch prediction of the feasible region of VPP;

[0009] S4. Submit the predicted feasible region of VPP to the power market for optimal clearing and settlement;

[0010] S5. Notify each virtual power plant of the clearing result, and the virtual power plant executes power generation according to the plan.

[0011] In step S1, the objective function of the VPP operating cost minimization optimization model is as follows:

[0012]

[0013] In the formula, is the total cost of the virtual power plant; T is the set of all time scales; G is the set of all generators in the VPP; W is the set of all wind turbines in the VPP; S is the set of all photovoltaic generators in the VPP; and are the transmission power and the electricity purchase price of the boundary node respectively, and are the active power and the generation cost of the generator in the VPP respectively; and are the abandoned power and the penalty coefficient of the distributed wind power generation in the VPP respectively; and are the abandoned power and the penalty coefficient of the distributed photovoltaic in the VPP respectively.

[0014] The constraint conditions of the VPP operating cost minimization optimization model include:

[0015]

[0016]

[0017] In the formula, N and B are the sets of all nodes and branches of the VPP respectively, and are the active power injection vector and the reactive power injection vector of the virtual power plant node at time t respectively; and are the active power vector and the reactive power vector of the adjustable units of the virtual power plant at time t respectively; and are the predicted value vectors of the active power and the reactive power of the distributed photovoltaic at time t respectively; and are the predicted value vectors of the active power and the reactive power of the distributed wind turbine at time t respectively; and are the active power demand vector and the reactive power demand vector of the load at time t respectively; is the active power exchange of the virtual power plant boundary node in time period t; e g , e r , e w , e d and e o are the incidence matrices of the adjustable units, the distributed photovoltaic, the wind turbines, the load and the virtual power plant gateway node with each node respectively; U i,t and U j,tis the square of the voltage magnitude of node i and node j at time period t; r ij and x ij are the resistance and reactance of branch ij respectively; P ij,t 、Q ij,t and I ij,t are the active power, reactive power and the square of the current magnitude of branch ij at time period t respectively; and are the minimum and maximum values of the active power transmission of branch ij respectively; U min and U max are the minimum and maximum values of the square of the node voltage respectively; and are the minimum and maximum values of the current flowing through branch ij respectively; and are the minimum and maximum values of the active power output of generator g respectively; and are the minimum and maximum values of the reactive power output of generator g respectively.

[0018] In step S2, first use the equivalent projection method to define the multi-time period scheduling boundary of the virtual power plant, and then use the vertex search method to solve the VPP feasible region.

[0019] The multi-time period scheduling boundary of the virtual power plant is: for any coupling variable on the multi-time period scheduling boundary Ω,

[0020] there exists a set of variables x t that satisfy the variable operation feasible region Φ, specifically:

[0021]

[0022] In the formula, x t is an internal variable,

[0023] where,

[0024] In the formula, and are the active power transmission and the total cost of the boundary nodes of virtual power plant number v respectively;

[0025]

[0026] In the formula, U t is the vector of the square of the voltage of all nodes of the virtual power plant, and I t is the vector of the square of the current amplitude of all branches of the virtual power plant.

[0027] Using the vertex search method to solve the VPP feasible region includes:

[0028] 1) Determine the vertex search model

[0029]

[0030] In the formula, h is the objective function of vertex search, the value of vector α represents the direction of vertex search, and A and B are coupling variables and the internal variable x t coefficient matrices of, and c is the right-end constant coefficient vector;

[0031] The constraint conditions of the VPP feasible region also include the constraints related to the virtual power plant quotation, specifically:

[0032]

[0033] In the formula, M is a very large positive number;

[0034] 2) Initialization: Determine N w +1 initial vertices, where N w is the dimension of the coupling variable; let α T = ±e i , e i is the standard basis vector with the i-th element being 1 and other elements being 0; store the initial vertices in the initialized vertex set V (0) ;

[0035] 3) Inner loop: The convex hull formed by the existing vertices generated in the k-th loop is Φ (k) , assume Φ (k) is surrounded by j planes, and the new vertex searches along the outer normal vector direction of the plane of Φ (k) ; the outer normal vector of the j-th plane is Then the newly identified vertex is to make and solve the optimal solution of the vertex search model, specifically:

[0036]

[0037] In the formula, △h (k,j) is the improvement ratio of the new vertex to the current convex hull, is the outer normal vector of the j-th plane, and d (k,j) is the right-side constant term of the corresponding hyperplane equation;

[0038] 4) Outer loop: Use the Hausdorff distance to evaluate the error D (k) between the current convex hull and the true projection:

[0039]

[0040] When D(k) When it is less than the allowable error, terminate and output the convex hull Φ composed of the existing nodes. (k) As the final result, otherwise enter the next outer loop.

[0041] Train the KAN network to achieve batch prediction of the VPP feasible region, including:

[0042] The input of the KAN network is a set of system parameters, including system load, wind power, and photovoltaic data. The output of the KAN network is the coordinate points of the predicted virtual power plant feasible region.

[0043] The power market operator performs optimal clearing based on the submitted VPP feasible region and market demand, including:

[0044] The objective function of the main-distribution-micro collaborative power market clearing is:

[0045]

[0046] In the formula, V is the set of all virtual power plants VPP in the system; λ g , c w , c s are the unit output cost coefficient and the penalty cost coefficients for abandoned wind and light respectively; P g,t , P c,w,t and P c,s,t are the output of the generator unit, the amount of abandoned wind power, and the amount of abandoned photovoltaic power respectively. is the quotation of the virtual power plant VPP at time t;

[0047] The constraint conditions are:

[0048]

[0049] In the formula, R is the set of all nodes in the system; P w,t and P s,t are the actual wind power and photovoltaic power outputs respectively; G g,r , W w,r , S s,r are the incidence matrices between the unit, wind farm, photovoltaic power station, and nodes respectively; D r,t is the load demand of the node; is the net injection power of the node; B r,n is the admittance matrix element of the branch between nodes r and n; θ r,t and θ n,t are the phase angles of nodes r and n; P g,t and P g,t+1 are the active power outputs of the generator unit at times t and t + 1; μ g,t and μ g,t+1 are the start-stop variables of the generator unit at times t and t + 1; μ g,kis the start-stop variable of unit g at time k; and are the minimum and maximum active power outputs of generator g, respectively; and are the upward and downward ramping rates of generator g, respectively; and are the minimum on-time and minimum off-time of generator g, respectively; k on and k off are auxiliary variables that constitute the minimum start-stop time constraint.

[0050] Let the operation feasible region obtained for this set of system parameters be: Submit the VPP feasible region information to the power market, and the power market operator forms the relevant constraints as:

[0051]

[0052] where x · and y · are the coordinate parameter values of the corresponding point, is the electricity purchase quantity of the VPP, is the total cost of the VPP; perform optimal clearing according to the relevant constraints of the power market operator to obtain the optimal and

[0053] The virtual power plant participation market rapid quotation generation system under the main-distribution-micro coordination includes:

[0054] A cost module for establishing an optimization model for minimizing the operating cost of the VPP;

[0055] A solution module for using the vertex search method to solve the VPP feasible region and form training data;

[0056] A prediction module for training the KAN network to achieve batch prediction of the VPP feasible region;

[0057] A settlement module for submitting the predicted VPP feasible region to the power market for optimal clearing and settlement;

[0058] An execution module for notifying each virtual power plant of the clearing result, and the virtual power plant executes power generation according to the plan.

[0059] In the operation of the present invention, by introducing the equivalent projection method and deep learning technology, it can effectively accelerate the quotation and clearing process of the virtual power plant participating in the power market under the main-distribution-micro coordination, solve the problems of computational efficiency and privacy protection in traditional methods, and improve the flexibility and economy of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 is the process of the vertex search method

[0061] Figure 2 is the comparison diagram of the KAN and MLP structures

[0062] Figure 3 is the comparison diagram of the prediction effects of KAN and MLP

[0063] Figure 4 is the display diagram of the acceleration effect of the KAN network prediction under different time scales

[0064] Figure 5 is the method flow chart of the present invention Specific implementation manner

[0065] As Figure 5 shown, the method for the virtual power plant to participate in the market rapid quotation generation under the main and auxiliary micro coordination includes the following steps:

[0066] S1. Establish an optimization model for minimizing the operating cost of the VPP;

[0067] S2. On the basis of step S1, use the vertex search method to solve the feasible region of the VPP (the full name is the operational feasible region, abbreviated as the feasible region) to form training data.

[0068] S3. Train the KAN network to achieve batch prediction of the VPP feasible region. Through the learnable activation function, KAN can better capture the complex patterns and relationships in the data.

[0069] S4. Submit the predicted VPP feasible region to the power market for optimal clearing and settlement, and transform the objective function and constraint conditions of the VPP into the constraint conditions of the coupling variables and submit them to the power market for optimal clearing, which can avoid the slow convergence problem of the iterative calculation method.

[0070] S5. Notify each virtual power plant of the clearing result, and the virtual power plant executes power generation according to the plan.

[0071] In the present invention, the dispatching and clearing of the virtual power plant participating in the power market meet the rapidity requirements to ensure that all participants have enough time to prepare and respond, thereby improving the market efficiency, facilitating timely adjustment to cope with the uncertain factors in the power system, and improving the flexibility and adaptability of the system.

[0072] In step S1, the objective function of the optimization model for minimizing the operating cost of the VPP is:

[0073]

[0074] In the formula, is the total cost of the virtual power plant; T is the set of all time scales; G is the set of all generators in the VPP; W is the set of all wind turbines in the VPP; S is the set of all photovoltaic generators in the VPP; and are the transmission power and the electricity purchase price of the boundary node respectively, and are the active power and the generation cost of the generators in the VPP respectively; and are the abandoned power and the penalty coefficient of distributed wind power generation in the VPP respectively; and are the abandoned power and the penalty coefficient of distributed photovoltaics in the VPP respectively.

[0075] The constraint conditions of the VPP operation cost minimum optimization model include:

[0076]

[0077] In the formula, N and B are the sets of all nodes and branches of the VPP respectively, and are the active power injection vector and the reactive power injection vector of the virtual power plant node at time t respectively; and are the active power vector and the reactive power vector of the adjustable units of the virtual power plant at time t respectively; and are the predicted value vectors of the active power and the reactive power of distributed photovoltaics at time t. Among them, in the present invention, distributed new energy is set to operate at a constant power factor, that is, the reactive power of new energy ( is the constant power factor vector of new energy); and are the predicted value vectors of the active power and the reactive power of distributed wind turbine sets at time t respectively; and are the active power demand vector and the reactive power demand vector of the load at time t respectively; is the exchanged active power of the virtual power plant boundary node in time period t; e g 、e r 、e w 、e d and e o are the incidence matrices of the adjustable units, distributed photovoltaics, wind turbines, loads and the virtual power plant gateway node with each node respectively; U i,t and U j,t are the squares of the voltage amplitudes of node i and node j at time t; r ij and x ij are the resistance and reactance of branch ij respectively; P ij,t, Q ij,t and I ij,t are respectively the square of the active power, reactive power and current amplitude of branch ij during period t; and are respectively the minimum and maximum values of the active power transmission power of branch ij; U min and U max are respectively the minimum and maximum values of the square of the node voltage; and are respectively the minimum and maximum values of the current flowing through branch ij; and are respectively the minimum and maximum values of the active power output of generator g; and are respectively the minimum and maximum values of the reactive power output of generator g.

[0078] In step S2, first use the equivalent projection method to define the multi-period scheduling boundary of the virtual power plant, and then use the vertex search method to solve the feasible region of the VPP.

[0079] The equivalent projection method is the process of calculating the operation feasible region of the virtual power plant, which is to project the high-dimensional space onto the low-dimensional space. The vertex search method is the method of calculating the operation feasible region.

[0080] Considering that the long-distance transmission of reactive power in the power grid will affect the voltage quality and cause an increase in active power loss, the reactive power in the power grid often adopts the method of local balance. Therefore, this invention studies in the mode that the boundary nodes of the virtual power plant only exchange active power with the main grid, and temporarily does not consider the reactive power exchange. Then the coupling variable of the virtual power plant is the active power exchanged by the boundary nodes. The multi-period scheduling of the virtual power plant needs to satisfy the VPP node power balance constraints (1-2), (1-3), branch power flow constraints (1-4), (1-5), system state variable constraints (1-6)-(1-8), adjustable unit output upper and lower limit constraints (1-9), (1-10).

[0081] The algorithms for the virtual power plant to participate in the optimal clearing of the power market mainly include centralized optimization algorithms and distributed optimization algorithms. The centralized optimization algorithm has many variables, slow solution, and requires collecting the device parameter information of the virtual power plant, which involves the problem of privacy protection. The distributed algorithm can effectively protect the device privacy of the virtual power plant, but it needs to be solved through multiple iterations and has difficulty in convergence.

[0082] Equivalent Projection (EP) is a technique used for optimization and modeling, mainly for converting complex multi-layer optimization problems into simpler single-layer problems. Its core idea is to "project" the internal variables and constraint conditions of a system into a low-dimensional coordinated variable space through mathematical transformation, thereby simplifying the scale and complexity of the problem while retaining the essential characteristics of the system.

[0083] Equivalent Projection maps the operation feasible region from a high-dimensional space (including internal variables) to a low-dimensional space (only including external variables) through mathematical methods. In the low-dimensional space, the optimization problem becomes simpler and can be solved directly. The equivalent projection method converts the solution of the optimization problem from a complex multi-layer structure to a single-layer structure, thus avoiding the complexity and computational burden of iterative calculations.

[0084] The variables related to the operation scheduling of the virtual power plant are defined as two categories:

[0085] 1) Coupled variables (i.e., coordinated variables) The active power exchange and quotation of the boundary nodes of the virtual power plant in each period, that is:

[0086] Among them, and are the active power transmission power and total cost of the boundary nodes of virtual power plant number v respectively, where v represents the number of the virtual power plant. The value of this variable can be obtained through the optimal clearing of the power market and used as the boundary condition for the optimization decision of the virtual power plant to fix the variable.

[0087] 2) Internal variable x t : Among them, U t is the vector of the square of the voltage magnitude of all nodes of the virtual power plant, and I t is the vector of the square of the amplitude of the current of all branches of the virtual power plant. For fixed coupled variables the internal variable x t can be determined by the self-scheduling optimization of the virtual power plant, that is, given the internal variable x t is obtained by solving (1-1)-(1-10).

[0088] Let the variable operation feasible region that satisfies the multi-period scheduling operation constraints (1-1)-(1-10) of the virtual power plant be Φ, and define the multi-period scheduling boundary of the virtual power plant as: for any coupled variable on the multi-period scheduling boundary Ω, there exists a set of feasible control and state variables x t that satisfy the variable operation feasible region Φ, that is, equation (1-11):

[0089]

[0090] The operating feasible region Φ refers to the set of all possible states in which the system can operate under given constraints. The virtual power plant calculates its operation feasible region through the equivalent projection method and submits this information to the power market for optimal clearing, thus avoiding complex distributed optimization and iterative calculations.

[0091] Solving the VPP feasible region using the vertex search method includes:

[0092] 1) Determining the vertex search model

[0093]

[0094] In the formula, h is the objective function of vertex search, the value of vector α represents the direction of vertex search, A and B are the coefficient matrices of the coupling variable and the internal variable x t respectively, and c is the right-end constant coefficient vector.

[0095] For the solution of the operating feasible region of the virtual power plant, its constraints are constraints (1-2)-(1-10), and the constraint (1-13) related to the virtual power plant's quotation, where M is a very large positive number.

[0096]

[0097] 2) Initialization: At least N w +1 initial vertices need to be determined, where N w is the dimension of the coupling variable. These initial vertices can be determined along the axis, that is, solving (1-12), and letting α T =±e i , e i is the standard basis vector with the i-th element being 1 and other elements being 0. Store the initial vertices in the initialized vertex set V (0) .

[0098] 3) Inner loop: The inner loop of the vertex search method constructs the convex hull formed by the existing vertices and identifies new vertices located outside the convex hull. The convex hull formed by the existing vertices generated in the k-th loop is Φ (k) , assuming that Φ (k) is surrounded by j planes, and the new vertices are searched along the outer normal vector direction of the planes of Φ (k) . The outer normal vector of the j-th plane is Then the newly identified vertex is the optimal solution of solving equation (1-12) after letting as follows:

[0099]

[0100] Calculate the distance from the vertex by formula (1-14). to the plane of Φ (k) , which is always non-negative. In the formula, △h (k,j) is the improvement ratio of the new vertex to the current convex hull, is the outer normal vector of the j-th plane, and d (k,j) is the right-side constant term of the corresponding hyperplane equation. If △h (k,j) >0, it indicates that is outside Φ (k) , add it to the vertex set V (k) , making Φ (k) closer to the actual feasible region and record the improvement ratio.

[0101] 4) Outer loop: The outer loop evaluates the error of the current approximation and compares it with the allowable error to decide whether to terminate the algorithm. Use the Hausdorff distance to evaluate the error D (k) between the current convex hull and the true projection:

[0102] When D (k) is less than the allowable error, the vertex search method terminates and outputs the convex hull Φ (k)

[0103]

[0104] composed of the existing nodes as the final result, otherwise enter the next outer loop.

[0105] As Figure 1 shown, the vertex search method first obtains four initial vertices V (0) through initialization, discovers new vertices (2.83, 113.35) and (2.27, 101.6) through the inner loop, calculates the error through the outer loop to terminate the loop process, and outputs the convex hull composed of the feasible region as the final result.

[0106] Train the KAN network to achieve batch prediction of the VPP feasible region, including:

[0107] The variable parameters of the system at a certain moment are mainly the system load parameter P, the wind power coefficient W, and the photovoltaic coefficient S. First, generate multiple groups of random system parameter sets {P, W, S}. For each group of system parameters, use the vertex search method to solve its corresponding operating feasible region. Use several vertices V = {(x, y)} at the bottom of the operating feasible region polygon to represent the operating feasible region. Multiple groups of random system parameter sets {P, W, S} are the training input set, and the operating feasible region set V is the training output set. Use these two data sets to train the deep neural network.

[0108] Due to the large difference in the ranges of the X - coordinate and Y - coordinate, the prediction effect of using a single neural network to predict the coordinates V of the bottom - edge vertices of the operation feasible region is poor. Two sets of neural networks are tried to predict the horizontal and vertical coordinates respectively to improve the prediction accuracy. It is found that the neural network can predict the X - coordinate well, but can only barely accurately predict the Y - coordinate. Therefore, in this invention, only the X - coordinate is predicted, and the Y - coordinate is solved by an analytical method, that is, solving equations (1 - 1)-(1 - 10), and fixing the X - coordinate, let is the predicted value of the X - coordinate output by the neural network.

[0109] KAN is a novel neural - network architecture inspired by the Kolmogorov - Arnold representation theorem. Different from the traditional multi - layer perceptron (MLP), KAN uses learnable activation functions on the edges of the network instead of fixed activation functions on the nodes, such as Figure 2 shown. This design allows each weight parameter in KAN to be replaced by a univariate function, usually parameterized in the form of a spline function, thus providing extremely high flexibility and being able to simulate complex functions with fewer parameters, enhancing the interpretability of the model.

[0110] For a general KAN network with L layers and an input vector The output of KAN is:[[]]

[0111]

[0112] where, Φ l is the function matrix corresponding to the l - th layer, x 0 is the input variable, and KAN(x 0 ) is the output variable of the KAN network.

[0113]

[0114] where, x l is the input variable of the l - th layer, and the activation function φ(x) is a combination of the basis function b(x) and the spline function spine(x):

[0115] φ(x)=ω b b(x)+ω s spine(x) (1 - 18)

[0116] where, ω b and ω s are the weight parameters related to the activation function, used to control the contribution ratios of the basis function b(x) and the spline function spine(x) in the overall activation function φ(x). In this invention, the basis function is set to silu(x), that is:

[0117]

[0118] Typically, the spline function spine(x) is represented as a linear combination of B-splines B(x):

[0119]

[0120] where c i are trainable parameters. The KAN network can be applied to a variety of regression and classification problems, especially when dealing with high-dimensional data and non-linear feature extraction, it can better capture the complex patterns and relationships in the data.

[0121] In summary, the KAN network is a new type of neural network architecture, characterized by having learnable activation functions on the edges of the network, rather than using fixed activation functions at the nodes like traditional MLPs. The KAN network consists of L layers, each with an input and an output. Each layer has a function matrix Φ l , which contains the activation functions of that layer. The activation function φ(x) in the KAN network is a combination of a basis function b(x) and a spline function. The spline function is usually parameterized as a linear combination of B-splines, where the parameter c i is trainable.

[0122] In the present invention, the input of the KAN network is a set of system parameters, mainly including system load, wind power, and photovoltaic data. These input data represent the state of the power system at a specific time and serve as the basis for predicting the operating feasible region (OFR) of the virtual power plant. The output of the KAN network is the coordinate points of the predicted virtual power plant OFR. These coordinate points define the boundaries within which the virtual power plant can operate in the power market and will serve as the constraints for the power market clearing process.

[0123] The power market operator performs optimized clearing based on the submitted VPP feasible region and market demand, including:

[0124] The virtual power plant aggregates a large number of dispersed distributed energy resources (DERs) such as photovoltaic, wind power, and adjustable units in the distribution network, calculates its operating feasible region, and reports it to the main grid. The operating feasible region, that is, the virtual power plant dispatching boundary, refers to the adjustable range of the variables between the virtual power plant and the main grid and the range that does not violate the operating constraints. The virtual power plant dispatching boundary is essentially the dimensionality reduction mapping of the high-dimensional feasible region of the virtual power plant dispatching model in the coupling variable space of the virtual power plant boundary nodes, that is, for each point in the operating feasible region, the virtual power plant can find a set of dispatching decisions that do not violate its own system steady-state constraints.

[0125] The virtual power plant submits the VPP feasible region to the main grid and participates in the optimal dispatching of the power market with this. The power market issues dispatching instructions to the virtual power plant, and then the virtual power plant conducts the final decision-making calculation.

[0126] The equivalent projection method aggregates a large amount of DER, and since the virtual power plant only needs to submit the operation feasible region, it effectively protects the data privacy of the VPP.

[0127] Generally, the objective function of the power market clearing considering the coordination of the main, distribution, and micro-grids can be expressed as:

[0128]

[0129] In the formula, V is the set of all virtual power plants VPP in the system; λ g , c w , c s are respectively the unit output cost coefficient and the penalty cost coefficients for abandoned wind and light; P g,t , P c,w,t and P c,s,t are respectively the output of the generator unit, the amount of abandoned wind power, and the amount of abandoned photovoltaic power. is the quotation of the virtual power plant VPP at time t.

[0130] The main constraint conditions are:

[0131]

[0132] In the formula, R is the set of all nodes in the system; P w,t and P s,t are respectively the actual wind power and photovoltaic power output; G g,r , W w,r , S s,r are respectively the incidence matrices between the unit, the wind farm, the photovoltaic power station, and the node; D r,t is the load demand of the node; is the net injection power of the node; B r,n is the admittance matrix element of the branch between nodes r and n; θ r,t and θ n,t are the phase angles of nodes r and n; P g,t and P g,t+1 are the active power outputs of the generator unit at times t and t + 1; μ g,t and μ g,t+1 are the start-stop variables of the generator unit at times t and t + 1; μ g,k is the start-stop variable of unit g at time k; and are respectively the minimum and maximum values of the active power output of generator g; and They are the upward and downward ramping rates of generator g, respectively; and They are the minimum up-time and minimum down-time of generator g, respectively; k on and k off are auxiliary variables that constitute the minimum start-stop time constraint.

[0133] Constraint (1-22) is the nodal power balance constraint of the system, (1-23) is the power flow constraint, (1-24) is the upper and lower limits constraint of the active power output of the generator set, (1-25) and (1-26) are the ramping constraints of the generator set, and (1-27)-(1-30) are the minimum start-stop time constraints of the generator set. Due to the large number of binary variables μ g,t and constraints (1-24)-(1-28), this problem becomes a mixed-integer linear programming problem, and its solution time increases significantly with the increase of the problem scale.

[0134] Use the trained neural network to predict the coordinate points of the operating feasible region. Assume that for this set of system parameters, the obtained operating feasible region is: Submit the VPP feasible region information to the power market, and the power market operator forms the relevant constraints as:

[0135]

[0136] where, x · and y · are the coordinate parameter values of the corresponding points (i.e., corresponding to x A and x A , x B and x B , x C and x C ), is the electricity purchase quantity of the VPP, is the total cost of the VPP. The power market optimizes and clears according to constraints (1-31)-(1-33) to obtain the optimal and The "^" in the variables represents the optimal value for easy distinction.

[0137] Each VPP executes power generation according to the optimized clearing result plan, including: according to the optimized clearing result of the power market and Each VPP executes power generation according to the optimized clearing result plan, that is:

[0138]

[0139] And satisfy: and To optimize the optimal electricity purchase quantity obtained from clearing and the quotation cost of the VPP, and are the electricity purchase quantity and the total cost of the VPP respectively. Each VPP executes the power generation plan according to the optimized clearing result, and obtains the results of all other decision variables.

[0140] The virtual power plant participation market rapid quotation generation system under the main and distribution micro coordination includes:

[0141] A cost module for establishing an optimization model for minimizing the operating cost of the VPP;

[0142] A solution module for using the vertex search method to solve the feasible region of the VPP and form training data;

[0143] A prediction module for training the KAN network to achieve batch prediction of the feasible region of the VPP;

[0144] A settlement module for submitting the predicted feasible region of the VPP to the power market for optimal clearing and settlement;

[0145] An execution module for notifying each virtual power plant of the clearing result, and the virtual power plant executes power generation according to the plan.

[0146] The present invention solves the operating feasible region of the virtual power plant through the equivalent projection method, and realizes the integrated optimization of the operating states and resources of different power grid levels. The objective function and constraint conditions of the virtual power plant are transformed into the constraint conditions of the coordination variables and submitted to the power market for optimal clearing. This transformation not only avoids the slow convergence problem of the traditional iterative calculation method, but also promotes the information sharing and decision coordination among the power grids at all levels under the main and distribution micro coordination, and improves the operating efficiency of the entire power system. By using the KAN (Kolmogorov - Arnold Networks) network, the efficient batch prediction of the operating feasible region of the virtual power plant is realized, and the problem of low efficiency in separately calculating the operating feasible region is solved.

[0147] Specific example

[0148] 1. Comparison of KAN acceleration effect at a single moment

[0149] For the virtual power plant participating in the optimal clearing of the power market at a single moment, Φ 1 and Φ 2 are submitted to the market operator through constraints (1 - 31)-(1 - 33). After obtaining the tie - line power, it is allocated to each VPP for decision - making calculation, and the results are shown in Table 1:

[0150] Table 1 Comparison of solution results at a single moment

[0151]

[0152] FromFigure 3 It can be seen that the prediction results of KAN and MLP are very close. Among them, the predicted feasible region of VPP2 using both methods almost completely coincides with the theoretical value, and the KAN network performs better in predicting the feasible region of VPP1. As can be seen from Table 1, although the prediction effects of using both methods in predicting the feasible region of VPP are close, using the KAN network for prediction can achieve a better objective function value and a more obvious acceleration effect.

[0153] 2. Comparison of the solution times for the day-ahead market optimization and clearing of large-scale power systems under the coordination of the main and distribution networks and microgrids

[0154] The IEEE 118-bus system is a relatively small system. Therefore, even for a time scale of 1440 hours, the solution times of both methods do not exceed 106 s, but it reflects the acceleration effect of the proposed framework. For larger systems, the solution effect will be more obvious. Taking the day-ahead scheduling optimization solution of the 13659-bus system as an example, there is one interval point every 15 minutes, with a total of 96 time points. The comparison of the solution results is shown in Table 2. When the objective function value only increases by 4.74%, the solution time is reduced by 64.40%, verifying the optimality and rapidity of the framework proposed in the present invention.

[0155] By reducing the problem scale, the equivalent projection method effectively speeds up the solution process, as Figure 4 shown. On the time scales of 720 hours and 1440 hours, compared with the centralized solution algorithm, the objective function values of the proposed solution method only increase by $227.68 and $1000.05 respectively, but the solution times are reduced by 26.92% and 43.45% respectively, which proves the rapidity and accuracy of the proposed method.

[0156] Aiming at the problem of slow calculation in the centralized and distributed optimization and clearing of virtual power plants participating in the power market under the coordination of the main and distribution networks and microgrids, the present invention proposes a quotation method based on the feasible region of VPP, uses the equivalent projection method to solve the feasible region of VPP, and uses the KAN network to predict the feasible region of VPP to avoid the iterative calculation of the distributed algorithm, realizing the batch characterization of the feasible region of VPP and achieving efficient and accurate calculation of VPP participating in the power market optimization and clearing. It is verified on the main network of 13659 buses and two IEEE-33 bus VPP test systems. The objective function value of the solution result only increases by 4.74%, but the solution time is reduced by 64.40%, reflecting the accuracy and rapidity of the VPP quotation method and the market clearing method proposed in the present invention.

[0157] Table 2 Comparison of the day-ahead scheduling solution results on a 24-hour time scale

[0158]

[0159] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered by the scope of the claims of the present invention.

Claims

1. A method for generating rapid quotation for virtual power plant participation in the market under the coordination of main power distribution and micro-power, characterized in that: The following steps are involved: S1. Establish a VPP operation cost minimization optimization model; S2, based on step S1, using the vertex search method to solve the VPP feasible domain to form training data; S3, train the KAN network to achieve batch prediction of the VPP feasible domain; S4. Submit the predicted VPP feasible domain to the power market for optimal clearing and settlement; S5. Notify each virtual power plant of the clearing results, and the virtual power plant will generate electricity as planned.

2. The method for generating fast quotation for virtual power plant participation in the market under the coordination of main power distribution and micro power generation according to claim 1 is characterized in that: In step S1, the objective function of the VPP operation cost minimization optimization model is: In the formula, is the total cost of the virtual power plant; T is the set of all time scales; G is the set of all generators in the VPP; W is the set of all wind turbines in the VPP; S is the set of all photovoltaic generators in the VPP; and are the transmission power and electricity purchase price of the boundary nodes respectively, and are the active power and generation cost of the generators in the VPP respectively; and are the abandoned power and penalty coefficient of distributed wind power generation in VPP respectively; and They are the abandoned power and penalty coefficient of distributed photovoltaic in VPP respectively.

3. The method for generating fast quotation for virtual power plant participation in the market under the coordination of main power distribution and micro power generation according to claim 2 is characterized in that: The constraints of the VPP operation cost minimization optimization model include: Where N and B are the sets of all nodes and branches of VPP, and They are the active power vector and reactive power vector injected into the virtual power plant node at time t respectively; and They are respectively the active power vector and reactive power vector of the adjustable units of the virtual power plant at time t; and are the predicted value vectors of distributed photovoltaic active power and reactive power at time t respectively; and are respectively the predicted value vectors of active power and reactive power of distributed wind turbine generator set at time t; and They are the load active power demand vector and reactive power demand vector at time t respectively; The active power exchanged by the boundary nodes of the virtual power plant during period t; e g 、e r 、e w 、e d and e o are the association matrices of adjustable units, distributed photovoltaics, wind turbines, loads, virtual power plant gateway nodes and each node; U i,t and U j,t is the square of the voltage amplitude between nodes i and j in time period t; r ij and x ij are the resistance and reactance of branch ij respectively; P ij,t , Q ij,t and I ij,t are respectively the active power, reactive power and square of current amplitude of branch ij in time period t; and are the minimum and maximum active transmission power of branch ij respectively; U min and U max is the minimum and maximum value of the square of the node voltage; and are the minimum and maximum values ​​of the current flowing through branch ij respectively; and are the minimum and maximum values ​​of the active output power of generator g respectively; and are the minimum and maximum values ​​of the reactive output power of generator g respectively.

4. The method for generating fast quotation for virtual power plant participation in the market under the coordination of main power distribution and micro power generation according to claim 1 is characterized in that: In step S2, the equivalent projection method is first used to define the multi-period dispatch boundary of the virtual power plant, and then the vertex search method is used to solve the feasible domain of the VPP.

5. The method for generating fast quotation for virtual power plant participation in the market under the coordination of main power distribution and micro power generation according to claim 4 is characterized in that: The multi-period dispatch boundary of the virtual power plant is: For any coupling variable on the multi-period dispatch boundary Ω The value of There exists a set of variables x t Satisfy the variable operation feasible domain Φ, specifically: In the formula, x t is an internal variable, in, In the formula, and are the active transmission power and total cost of the boundary nodes of the virtual power plant number v, respectively; Where U t is the square vector of voltage of all nodes in the virtual power plant, I t It is the vector of the squared current amplitudes of all branches of the virtual power plant.

6. The method for generating fast quotation for virtual power plant participation in the market under the coordination of main power distribution and micro power generation according to claim 5 is characterized in that: Use the vertex search method to solve the VPP feasible region, including: 1) Determine the vertex search model Where h is the objective function of vertex search, the value of the α vector represents the direction of vertex search, A and B The coupling variables are and the internal variable x t The coefficient matrix of , c is the constant coefficient vector on the right side; The constraints of the VPP feasible region also include constraints related to virtual power plant quotations, specifically: In the formula, M is a very large positive number; 2) Initialization: Determine N w +1 initial vertex, where N w is the dimension of the coupled variable; let α T = ±e i , e i is a standard basis vector with the i-th element being 1 and the other elements being 0; the initial vertex is stored in the initialization vertex set V (0) middle; 3) Inner loop: The convex hull of the existing vertices generated by the kth loop is Φ (k) , let Φ (k) Surrounded by j planes, the new vertex is along Φ (k) The external normal vector direction of the plane is searched; the external normal vector of the jth plane is Then the newly identified vertex For order Then solve the optimal solution of the vertex search model, specifically: In the formula, △h (k,j) For the new vertex The improvement ratio to the current convex hull, is the external normal vector of the jth plane, d (k,j) is the constant term on the right side of the corresponding hyperplane equation; 4) Outer loop: Use Hausdorff distance to evaluate the error D between the current convex hull and the true projection (k) : When D (k) When it is less than the allowable error, terminate and output the convex hull Φ composed of the existing nodes (k) As the final result, otherwise enter the next outer loop.

7. The method for generating fast quotation for virtual power plant participation in the market under the coordination of main power distribution and micro power generation according to claim 1 is characterized in that: Training the KAN network to achieve batch prediction of the VPP feasible domain includes: The input of the KAN network is a set of system parameters, including system load, wind power and photovoltaic data, and the output of the KAN network is the coordinate points of the predicted feasible domain of the virtual power plant.

8. The method for generating fast quotation for virtual power plant participation in the market under the coordination of main power distribution and micro power generation according to claim 1 is characterized in that: The electricity market operator optimizes and clears according to the submitted VPP feasible domain and market demand, including: The objective function of power market clearing for the main distribution and micro-cooperation is: Where V is the set of all virtual power plants VPP in the system; g 、c w 、c s are the unit output cost coefficient and the penalty cost coefficient for abandoning wind and solar power; P g,t , P c,w,t and P c,s,t They are generator unit output, wind power abandonment and photovoltaic abandonment. is the quotation of the virtual power plant VPP in time period t; The constraints are: Where R is the set of all nodes in the system; P w,t and P s,t Actual wind power and photovoltaic output respectively; G g,r , W w,r , S s,r are the association matrices between units, wind farms, photovoltaic power stations and nodes; D r,t is the load demand of the node; The net injected power of the node; B r,n is the admittance matrix element of node r and branch n; θ r,t and θ n,t is the phase angle between nodes r and n; P g,t and P g,t+1 is the active output of the generator set at time t and t+1; μ g,t and μ g,t+1 is the start and stop variable of the generator set at time t and t+1; μ g,k is the start and stop variable of unit g at time k; and are the minimum and maximum active output of generator g respectively; and are the ramp-up and ramp-down rates of generator g, respectively; and They are the minimum start time and minimum stop time of generator g; k on and k off is the auxiliary variable that constitutes the minimum start-stop time constraint.

9. The method for generating fast quotation for virtual power plant participation in the market under the coordination of main power distribution and micro power generation according to claim 8 is characterized in that: Set The feasible domain of operation obtained by this set of system parameters is: Submit the VPP feasible domain information to the power market, and the power market operator forms the relevant constraints as follows: In the formula, x · and · is the coordinate parameter value of the corresponding point, is the amount of electricity purchased by VPP, is the total cost of the VPP; according to the relevant constraints of the power market operator, the optimal and 10. A system for generating rapid market quotations for virtual power plants under the coordination of main power distribution and micro-power, characterized in that: include: Cost module, used to establish the VPP operation cost minimization optimization model; A solution module, used to solve the feasible domain of VPP using vertex search method to form training data; Prediction module, used to train the KAN network to achieve batch prediction of the VPP feasible domain; The settlement module is used to submit the predicted VPP feasible domain to the power market for optimal clearing settlement; The execution module is used to notify each virtual power plant of the clearing results, and the virtual power plant executes power generation according to the plan.

Images