A multi-average field control method and device for a virtual power plant to participate in secondary frequency regulation

Abstract

The present invention relates to a multi-average field control method and device for a virtual power plant to participate in secondary frequency regulation, which includes the steps of: evaluating the adjustable space and comprehensive frequency regulation performance of heterogeneous distributed resources inside the virtual power plant by using a unified modeling method; dividing different types of distributed resources into multiple homogeneous clusters with the same response characteristics by using a dynamic granulation method; describing the dynamic interaction between a single homogeneous cluster and the average field term in the same type of distributed resources through the average field term and the HJB equation; and solving the equilibrium solution of the average field term and the HJB equation. This method is based on a multi-average field control framework model with interactive iteration between the HJB equation and the average field term, transforms the pairwise interaction between homogeneous DERs into the interaction between a single DER cluster and the average field composed of the remaining DERs, and decomposes the many-body problem into multiple single-body problems for distributed solution, realizing the rapid decomposition of regulation instructions inside the VPP and improving the VPP regulation speed.

Description

Technical Field

[0001] The present invention relates to the technical field of virtual power plant resource regulation, and particularly relates to a multi-average field control method and device for a virtual power plant to participate in secondary frequency modulation. Background Art

[0002] As a convergence point of energy supply and energy consumption, a virtual power plant (VPP) can aggregate distributed energy resources (DERs) to flexibly participate in the frequency modulation ancillary service market. When the VPP participates in the ancillary service market, it needs to declare information such as a bid price curve and adjustable capacity to the power grid in advance. Existing research only models scenarios under the aggregation of certain specific resources (such as energy storage, users, etc.), resulting in the particularity of the model and poor scalability of the model itself, which brings difficulties to the formulation of the operation strategy for the virtual power plant operator (VPPO) to flexibly aggregate different types of resources in advance. In addition, the frequency modulation ancillary service market puts forward high requirements for the instruction decomposition speed of the VPP: on the one hand, since the DERs aggregated by the VPP usually belong to different interest entities, the instruction decomposition within the VPP involves a large number of interactions between entities; on the other hand, while pursuing its own maximum benefit, the DERs need to respond to the regulation instructions of the VPP. In this case, there are a large number of interaction processes between DERs, but the computational complexity of the traditional control framework increases with the increase in the number of participants. When the number of participants is quite large, it may fall into the "curse of dimensionality" problem. Therefore, there is an urgent need to construct a reasonable and efficient control framework, which can evaluate the adjustable ability of the VPP according to different aggregation situations, formulate a scheduling strategy for the virtual power plant's advance bidding, and establish a more accurate regulation instruction decomposition model to reduce the scheduling error and ensure the maximization of individual benefits, so as to realize the rapid decomposition of the regulation instructions within the VPP. Summary of the Invention

[0003] To overcome the deficiencies of the above-mentioned prior art, the present invention provides a multi-average field control method and device for a virtual power plant to participate in secondary frequency modulation, and specifically adopts the following technical solutions:

[0004] A multi-average field control method for a virtual power plant to participate in secondary frequency modulation, which includes the following steps:

[0005] Evaluate the adjustable space and comprehensive frequency modulation performance of heterogeneous distributed resources inside the virtual power plant by using a unified modeling method;

[0006] Use a dynamic granulation method to divide different types of distributed resources into multiple homogeneous clusters with the same response characteristics;

[0007] Construct a multi-average field control framework model, and describe the dynamic interaction between a single homogeneous cluster and the average field term in the same type of distributed resources through the average field term and the HJB equation;

[0008] Solve the equilibrium solutions of the average field term and the HJB equation based on the distributed acceleration method.

[0009] Optionally: The steps of evaluating the adjustable space and comprehensive frequency regulation performance of heterogeneous distributed resources inside the virtual power plant by adopting the unified modeling method include:

[0010] According to the operating characteristics of heterogeneous distributed resources, use a convex set in the T-dimensional space to conduct unified modeling of the aggregation power feasible region of distributed resources:

[0011]

[0012] where u is the type of distributed resource, and the distributed resources include renewable energy, distributed energy storage, or temperature-controlled load; F u is the coefficient matrix, and where K represents the number of constraints within a time interval; b u is the constant matrix, and where T represents the number of scheduling time intervals; is the power of the i-th distributed resource of type u, is the vector of the charge of the i-th distributed resource of type u, where

[0013] Using the Minkowski sum, map to the aggregation power feasible region

[0014]

[0015] where n is the number of distributed resources; is the sum of point sets in the Euclidean space; is the aggregation power of the distributed resources of type u;

[0016] Use the embedded right-angled pyramid method to solve the operating power boundaries for different time periods to achieve time decoupling of the aggregated adjustable power:

[0017]

[0018] where and are the upper and lower bounds of the operating power; is the adjustable power interval of the distributed resources of type u;

[0019] Finally, three indicators, namely the adjustment rate, adjustment accuracy, and response duration, are selected to form a comprehensive indicator for evaluating the frequency regulation performance of distributed resources:

[0020]

[0021] Among them, is the comprehensive indicator; is the adjustment rate; is the adjustment accuracy; is the response duration; K i,1 is the adjustment rate of the single distributed resource i; K i,2 is the adjustment accuracy of the single distributed resource i; K i,3 is the response duration of the single distributed resource i.

[0022] Optionally: The step of using the dynamic granulation method to divide different types of distributed resources into multiple homogeneous clusters with the same response characteristics includes:

[0023] First, construct a four-dimensional information system containing detailed information of distributed resources:

[0024] S = (U, D, V, f);

[0025] where U = {DER u,1 , DER u,2 , …, DER u,n} is the universe of discourse, and DER u,i (i = 1, 2, …, n) represents the i-th distributed resource in the set of u-type distributed resources; D is the attribute set; V is the value set of attributes; f is the information function, which represents the mapping between each distributed resource and the value of the corresponding attribute;

[0026] Randomly aggregate the distributed resources in the universe of discourse U into cluster A u,c , where the conditional attributes of cluster A u,c are:

[0027]

[0028] Among them is the downward adjustable range of the aggregated power of cluster A u,c ; is the upward adjustable range of the aggregated power of cluster A u,c ; is the comprehensive frequency regulation performance index of cluster A u,c ;

[0029] Obtain the value set u,c of the attributes of cluster A

[0030]

[0031] According to cluster A u,c Obtain the value of each attribute according to the value set V of the attribute and the corresponding information function as follows:

[0032]

[0033] Where is the upper limit of the aggregated power of cluster A u,c ; is the lower limit of the aggregated power of cluster A u,c ; is the base value of the aggregated power of cluster A u,c ;

[0034] Determine whether there is an indistinguishable relationship between cluster A u,c and the standard cluster A u,s :

[0035]

[0036] Among them, Euc[f(A u,c , D), f(A u,s , D)] is the Euclidean distance between cluster A u,c and the standard cluster A u,s ; To meet the frequency modulation accuracy requirements, the constraints of the standard cluster A u,s are as follows

[0037]

[0038] Take the cluster A u,s that has an indistinguishable relationship with the standard cluster A u,c as the homogeneous cluster A u,d :

[0039]

[0040] Among them, E u,d is the charge of the homogeneous cluster A u,d ; is the upward regulation power of the homogeneous cluster A u,d ; is the downward regulation power of the homogeneous cluster A u,d ; is the upper limit of the upward regulation power of the standard cluster A u,s ; is the lower limit of the downward regulation power of the standard cluster A u,s ; is the upper limit of the charge of the standard cluster A u,s ; For the standard cluster A u,s The lower limit of the charge amount; For the homogeneous cluster A u,d The comprehensive frequency regulation performance index; For the homogeneous cluster A u,d The standard comprehensive frequency regulation performance index.

[0041] Optionally: The step of describing the dynamic interaction between a single homogeneous cluster and the mean-field term in the same type of distributed resources through the mean-field term and the HJB equation includes:

[0042] Based on the purpose of seeking optimal frequency regulation commands and economic benefit distribution, construct a dynamic stochastic control model according to the interaction between distributed resource clusters:

[0043]

[0044] Where is the set of participants, which represents the number of u-type distributed resource clusters participating in the t period, and is the state space, which represents the joint state of all clusters, and is the decision vector space, which represents the joint control strategy of all clusters, and {J u,d} is the utility function, which represents the preference degree of the participants in the interaction;

[0045] Based on the dynamic stochastic control model, obtain the revenue function of the homogeneous cluster A u,d :

[0046]

[0047] Where J u,d () is the revenue function; s u,d is the decision vector of the u-type distributed resource; S u,-d is the joint decision vector after excluding the homogeneous cluster A u,d from the u-type distributed resource cluster; For the homogeneous cluster A u,d The unit incentive price of the frequency regulation power; For the homogeneous cluster A u,d The frequency regulation power; Is the standard charge amount, μ u Is the penalty coefficient for deviating from the standard charge amount ; Is the standard frequency regulation power, And Are respectively the penalty coefficients for deviating from the standard frequency regulation power ;

[0048] Construct a corresponding dynamic incentive mechanism according to the regulation ability and frequency regulation demand of distributed resources:

[0049]

[0050] where is the unit incentive price of the frequency regulation power of the homogeneous cluster in renewable energy; is the unit incentive price of the frequency regulation power of the homogeneous cluster in distributed energy storage; is the unit incentive price of the frequency regulation power of the homogeneous cluster in the thermostatic load; h RES,d is a piecewise linear function describing the relationship between the regulation amount of renewable energy and the incentive price; h ESS,d is a piecewise linear function describing the relationship between the regulation amount of distributed energy storage and the incentive price; h TCL,d is a piecewise linear function describing the relationship between the regulation amount of thermostatic load and the incentive price; N RES N(t) is the number of renewable energy clusters participating in the t-th time period; N ESS N(t) is the number of distributed energy storage clusters participating in the t-th time period; N TCL N(t) is the number of thermostatic load clusters participating in the t-th time period; is a linear parameter reflecting the regulation amount of the homogeneous cluster A in the u-type distributed resource u,d ; is a linear parameter reflecting the influence of the regulation enthusiasm of the homogeneous cluster A in the u-type distributed resource u,d ; is the proportion of the total incentive obtained by renewable energy in the market clearing price, is the proportion of the total incentive obtained by distributed energy storage in the market clearing price; is the proportion of the total incentive obtained by thermostatic load in the market clearing price; f(t) is the system frequency at the t-th time period, and f0 is the rated frequency at the t-th time period; p FR p(t) is the market clearing price of the frequency regulation market at the t-th time period;

[0051] Combine the self-state of the homogeneous cluster A u,d and the dynamic incentive mechanism to obtain the first optimal control strategy for maximizing the frequency regulation benefit:

[0052]

[0053] where is the optimal control strategy at the equilibrium point in the first optimal control strategy;

[0054] Transform the first optimal control strategy based on the multi-average field control framework model to obtain the second optimal control strategy, and obtain the HJB equation of the multi-average field control framework model based on the second optimal control strategy.

[0055] Optionally: The mean - field term includes:

[0056]

[0057] where y RES (t) is the mean - field term describing the collective effect of RES; y ESS (t) is the mean - field term describing the collective effect of ESS; y TCL (t) is the mean - field term describing the collective effect of TCL; N RES (t) is the number of renewable - energy clusters participating in the t - time period; N ESS (t) is the number of distributed - energy - storage clusters participating in the t - time period; N TCL (t) is the number of temperature - controlled - load clusters participating in the t - time period; is the frequency - regulation power of a single cluster in the renewable energy participating in the t - time period; is the frequency - regulation power of a single cluster in the distributed energy storage participating in the t - time period; is the frequency - regulation power of a single cluster in the temperature - controlled load participating in the t - time period.

[0058] Optionally: The step of transforming the first optimal control strategy based on the multi - mean - field control - framework model to obtain the second optimal control strategy includes:

[0059] Obtain the relationship between the incentive price and the corresponding mean - field term in the homogeneous cluster according to the mean - field term and the dynamic incentive mechanism:

[0060]

[0061] where is the unit incentive price of the frequency - regulation power of the homogeneous cluster in the renewable energy; is the unit incentive price of the frequency - regulation power of the homogeneous cluster in the distributed energy storage; is the unit incentive price of the frequency - regulation power of the homogeneous cluster in the temperature - controlled load; y RES (t) is the mean - field term describing the collective effect of RES; y ESS (t) is the mean - field term describing the collective effect of ESS; y TCL (t) is the mean - field term describing the collective effect of TCL; h RES,d is the piece - wise linear function describing the relationship between the renewable - energy regulation amount and the incentive price; h ESS,d is the piece - wise linear function describing the relationship between the distributed - energy - storage regulation amount and the incentive price; h TCL,d is the piece - wise linear function describing the relationship between the temperature - controlled - load regulation amount and the incentive price; N RES (t) is the number of renewable - energy clusters participating in the t - time period; N ESS (t) is the number of distributed - energy - storage clusters participating in the t - time period; N TCL(t) is the number of thermostatic control load clusters participating in the t time period;

[0062] Optimize homogeneous cluster A based on the relationship between the incentive price and the mean field term in the homogeneous cluster u,d Revenue function of:

[0063]

[0064] where J u,d () is the revenue function; s u,d is the decision vector of the u - type distributed resource; y u is the mean field term describing the collective effect of the u - type DER; is the standard integrated frequency regulation performance index of homogeneous cluster A u,d ; h u,d is the piece - wise linear function describing the relationship between the regulation amount of the u - type distributed resource and the incentive price; N u (t) is the number of u - type distributed resource clusters participating in the t time period; y u (t) is the mean field term describing the collective effect of the u - type DER in the t time period; is the frequency regulation power of a single cluster among the u - type distributed resources participating in the t time period; is the standard frequency regulation power of a single cluster among the u - type distributed resources participating in the t time period; is the penalty coefficient for deviating from the standard frequency regulation power; μ u is the penalty coefficient for deviating from the standard charge amount; E u,d (t) is the charge amount of homogeneous cluster A in the t time period u,d ; is the standard charge amount of homogeneous cluster A in the t time period u,d ;

[0065] Optimize the first optimal control strategy based on the multi - mean - field control framework model to obtain the second optimal control strategy:

[0066]

[0067] Obtain the value function of the second optimal control strategy based on the Bellman optimal value principle:

[0068]

[0069] Based on the value function, further obtain the HJB equation of the second optimal control strategy under the multi - mean - field control framework model:

[0070]

[0071] where represents the operator for taking the partial derivative with respect to s u,d ; v u,d() is the value function of the optimal control strategy; t is the frequency regulation period; is the homogeneous cluster A u,d of the standard comprehensive frequency regulation performance index; h u,d is a piecewise linear function describing the relationship between the regulation amount and the incentive price of u-type distributed resources; N u (t) is the number of u-type distributed resource clusters participating in the t period; is the frequency regulation power of a single cluster among the u-type distributed resources participating in the t period; is the standard frequency regulation power of a single cluster among the u-type distributed resources participating in the t period; is the penalty coefficient for deviating from the standard frequency regulation power; μ u is the penalty coefficient for deviating from the standard charge; E u,d (t) is the charge of the homogeneous cluster A in the t period u,d ; is the homogeneous cluster A in the t period u,d of the standard charge.

[0072] Optionally: The steps of solving the equilibrium solution of the mean field term and the HJB equation include:

[0073] Select a solution algorithm and set the maximum number of iterations and the convergence parameter of the solution algorithm;

[0074] Combine the HJB equations of each homogeneous cluster A u,d with the multi-mean field control framework model to obtain the frequency regulation power of the homogeneous cluster A corresponding to the k-th iteration u,d respectively;

[0075] And calculate the mean field term of the homogeneous cluster A corresponding to the k-th iteration according to the frequency regulation power of the homogeneous cluster A corresponding to the k-th iteration u,d ; u,d of the homogeneous cluster A corresponding to the k-th iteration;

[0076] When the mean field term result of the homogeneous cluster A corresponding to the k-th iteration u,d satisfies the convergence parameter, output the frequency regulation power of the homogeneous cluster A corresponding to the k-th iteration u,d ; Otherwise, increment the current iteration number by 1 and calculate the frequency regulation power and the mean field term result of the homogeneous cluster A corresponding to the k+1-th iteration in turn. u,d

[0077] Optionally: The distributed acceleration algorithm adopts the Mann iteration algorithm:

[0078]

[0079] Among them, is the mean field term of the homogeneous cluster A corresponding to the k-th iteration u,d ;​ is the mean field term corresponding to the (k + 1)-th iteration for the homogeneous cluster A u,d ; is the coefficient sequence of the Mann iterative algorithm at the (k + 1)-th iteration; N u is the number of DER clusters of class u; is the frequency modulation power corresponding to the (k + 1)-th iteration for the homogeneous cluster A u,d ;

[0080] Furthermore, the present invention also discloses a multi-mean field control device for a virtual power plant to participate in secondary frequency modulation. The device includes:

[0081] A modeling and evaluation module, configured to evaluate the adjustable space and comprehensive frequency modulation performance of heterogeneous distributed resources inside the virtual power plant by using a unified modeling method;

[0082] A cluster division module, configured to divide different types of distributed resources into multiple homogeneous clusters with the same response characteristics respectively by using a dynamic granulation method;

[0083] A dynamic analysis module, configured to construct a multi-mean field control framework model, and describe the dynamic interaction between a single homogeneous cluster and the mean field term among the same type of distributed resources through the mean field term and the HJB equation;

[0084] A result solving module, configured to solve the equilibrium solution of the mean field term and the HJB equation based on a distributed acceleration method.

[0085] Beneficial effects

[0086] The technical solution of the present invention has obtained the following beneficial effects:

[0087] (1) The method of the present invention can be oriented to the frequency modulation auxiliary service market, uniformly model and evaluate the adjustable space of heterogeneous distributed resources, provide a basis for the VPP to declare the aggregated adjustable capacity on a daily basis, and effectively improve the scalability of the VPP to aggregate different types of DERs.

[0088] (2) The method of the present invention constructs a multi-mean field control framework model based on the interactive iteration between the HJB equation and the mean field term (MFT), transforms the pairwise interaction between homogeneous DERs into the interaction between a single DER cluster and the mean field composed of the remaining DERs, and decomposes the many-body problem into multiple single-body problems for distributed solution, realizing the rapid decomposition of the regulation instructions inside the VPP and improving the VPP regulation speed. Description of the drawings

[0089] Figure 1 is the control framework diagram inside the VPP in the embodiment of the present invention.

[0090] Figure 2Schematic diagram of the multi-average field control method for a virtual power plant to participate in secondary frequency regulation in an embodiment of the present invention.

[0091] Figure 3 Flowchart of the distributed acceleration algorithm in an embodiment of the present invention.

[0092] Figure 4 Schematic diagram of the valley filling power of different DER clusters under the optimal control strategy in an embodiment of the present invention.

[0093] Figure 5 Schematic diagram of the peak shaving power of different DER clusters under the optimal control strategy in an embodiment of the present invention.

[0094] Figure 6 Schematic diagram of the net income of different DER clusters in an embodiment of the present invention. Detailed implementation manners

[0095] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention and cannot be used to limit the protection scope of the present invention. It should be noted that the following detailed descriptions are all exemplary and are intended to provide further descriptions of the present application.

[0096] In the present invention, the distributed energy resources (DERs) aggregated by a virtual power plant (VPP) at least include renewable energy sources (RESs), distributed energy storage systems (ESSs), and temperature control loads (TCLs). When the above-mentioned distributed energy resources participate in the frequency regulation market (FRM), the control framework inside the VPP is as Figure 1 shown.

[0097] Combined with Figure 1 and Figure 2 shown, the present invention specifically discloses a multi-average field control method for a virtual power plant to participate in secondary frequency regulation, which includes the following steps:

[0098] Step 1: Evaluate the adjustable space and comprehensive frequency regulation performance of heterogeneous distributed energy resources inside the virtual power plant by using a unified modeling method;

[0099] Specifically, the specific process of evaluating the adjustable space and comprehensive frequency regulation performance of heterogeneous distributed energy resources inside the virtual power plant by using a unified modeling method in the first step includes:

[0100] First, according to the operating characteristics of heterogeneous distributed resources, a convex set in the T-dimensional space is used to uniformly model the aggregation power feasible region of distributed resources:

[0101]

[0102] where u is the type of distributed resource, and the distributed resources include renewable energy, distributed energy storage, or temperature-controlled load; F u is a coefficient matrix, and where K represents the number of constraints within a time interval; b u is a constant matrix, and where T represents the number of scheduling time intervals; is the power of the i-th distributed resource of type u, is the charge vector of the i-th distributed resource of type u, where

[0103] Using the Minkowski sum, is mapped to the aggregation power feasible region

[0104]

[0105] where n is the number of distributed resources; is the sum of point sets in the Euclidean space; is the aggregation power of the u-type distributed resources.

[0106] In the present invention, since the operation models of some DERs are time-varying, the aggregation power flexibility has time coupling, and geometrically, it can be abstracted as a high-dimensional aggregation power polyhedron, making it more difficult to obtain the exact operating power boundary. Therefore, in this step, the embedded right-angled pyramid method is used to solve the operating power boundaries in different time periods to achieve the time decoupling of the aggregatable adjustable power:

[0107]

[0108] where and are the upper and lower bounds of the operating power; is the adjustable power interval of the u-type distributed resources.

[0109] Finally, three indicators, namely the adjustment rate, adjustment accuracy, and response duration, are selected to form a comprehensive indicator for evaluating the frequency modulation performance of distributed resources:

[0110]

[0111] where, is the comprehensive indicator; For adjusting the rate; For adjusting the precision; For the response duration; K i,1 Is the adjustment rate of a single distributed resource i; K i,2 Is the adjustment precision of a single distributed resource i; K i,3 Is the response duration of a single distributed resource i.

[0112] Step 2: Use the dynamic granulation method to divide different types of distributed resources into multiple homogeneous clusters with the same response characteristics; In practical applications, a large number of DERs in VPP are rarely completely homogeneous, making it difficult to manage them uniformly. Therefore, in Step 2, the dynamic granulation method is used to divide large-scale DERs into smaller subsets with similar adjustment characteristics, namely DER clusters, for subsequent management; Specifically, the specific method of using the dynamic granulation method in Step 2 to divide different types of distributed resources into multiple homogeneous clusters with the same response characteristics includes:

[0113] First, construct a four-dimensional information system containing the detailed information of distributed resources. From the perspective of granular computing, the four-dimensional information system containing the detailed information of DERs can be defined as follows:

[0114] S = (U, D, V, f);

[0115] Where U = {DER u,1 , DER u,2 , …, DER u,n} is the universe of discourse, and DER u,i (i = 1, 2, …, n) represents the i-th distributed resource in the set of u-type distributed resources; D is the set of attributes; V is the set of values of the attributes, generally Where Is the value of attribute D m ; f is the information function, which represents the mapping between each distributed resource and the value of the corresponding attribute, that is, f: U × D → V.

[0116] Subsequently, randomly aggregate the distributed resources DERs in the universe of discourse U into cluster A u,c . Since the incentives obtained by the DER cluster should depend on its output and service quality in the frequency regulation ancillary service, the up and down adjustable power ranges define the conditional attributes of cluster A u,c :

[0117]

[0118] Where Is the downward adjustable range of the aggregated power of cluster A u,c ; Is the upward adjustable range of the aggregated power of cluster A u,c ; For Cluster A u,c 's comprehensive frequency regulation performance index;

[0119] Obtain the value set of the attributes of Cluster A according to the time decoupling method of aggregated adjustable power and the comprehensive index calculation method u,c

[0120]

[0121] Where is the upper limit of the aggregated power of Cluster A u,c ; is the lower limit of the aggregated power of Cluster A u,c ; is the comprehensive index of the u - type distributed resources, is the base value of the aggregated power of Cluster A u,c .

[0122] According to the value set V of the attributes of Cluster A u,c and the corresponding information function, the value of each attribute is:

[0123]

[0124] Judge whether there is an indistinguishable relationship between Cluster A u,c and the standard Cluster A u,s

[0125]

[0126] Where Euc[f(A u,c , D), f(A u,s , D)] is the Euclidean distance between Cluster A u,c and the standard Cluster A u,s ;

[0127] It should be noted that to meet the frequency regulation accuracy requirements, the constraints of the standard Cluster A u,s are as follows:

[0128]

[0129] Cluster A that has an indistinguishable relationship with the standard Cluster A u,s is regarded as the homogeneous Cluster A u,c , thus enabling a large number of heterogeneous DERs in the universe U to be divided into homogeneous DER clusters A u,d , where the homogeneous Cluster A u,d is expressed as: u,d

[0130] ​​​

[0131] where E u,d is the charge amount of the homogeneous cluster A u,d ; is the up-regulation power of the homogeneous cluster A u,d ; is the down-regulation power of the homogeneous cluster A u,d ; is the upper limit of the up-regulation power of the standard cluster A u,s ; is the lower limit of the down-regulation power of the standard cluster A u,s ; is the upper limit of the charge amount of the standard cluster A u,s ; is the lower limit of the charge amount of the standard cluster A u,s ; is the comprehensive frequency regulation performance index of the homogeneous cluster A u,d ; is the standard comprehensive frequency regulation performance index of the homogeneous cluster A u,d ;

[0132] Step 3: Construct a multi-average field control framework model, and describe the dynamic interaction between a single homogeneous cluster and the average field term in the same type of distributed resources through the average field term and the HJB equation;

[0133] Specifically, the process in which the average field term and the HJB equation in Step 3 describe the dynamic interaction between a single homogeneous cluster and the average field term in the same type of distributed resources includes:

[0134] The clearing results of various DERs in the FRM should be further allocated to the DER clusters. While pursuing their own maximum interests, the DER clusters are obliged to undertake the frequency regulation tasks of the VPP. In order to seek the optimal frequency regulation instructions and the distribution of economic benefits, interactions are formed among the DER clusters. Based on the purpose of seeking the optimal frequency regulation instructions and the distribution of economic benefits, a dynamic stochastic control model can be constructed according to the interactions among the distributed resource clusters:

[0135]

[0136] where is the set of participants, which represents the number of u-type distributed resource clusters participating at time t, and is the state space, which represents the joint state of all clusters, and is the decision vector space, which represents the joint control strategy of all clusters, and {J u,d} is the utility function, which represents the preference degree of the participants in the interaction;

[0137] Because each homogeneous cluster A u,dWhen pursuing maximum benefits while responding to the regulation instructions of the VPP, the utility function is defined as the net benefit. Therefore, based on the dynamic stochastic control model, the benefit function of homogeneous cluster A is obtained. u,d The benefit function of

[0138]

[0139] where J u,d () is the benefit function; s u,d is the decision vector of type-u distributed resources; S u,-d is the joint decision vector after excluding homogeneous cluster A u,d from the homogeneous cluster of type-u distributed resources; is the unit incentive price of the frequency regulation power of homogeneous cluster A u,d ; is the frequency regulation power of homogeneous cluster A u,d ; is the standard charge, and μ u is the penalty coefficient for deviating from the standard charge ; is the standard frequency regulation power, and are the penalty coefficients for deviating from the standard frequency regulation power respectively;

[0140] It should be noted that considering that homogeneous cluster A u,d has similar response characteristics, the standard frequency regulation power can be calculated from the clearing power as follows:

[0141]

[0142] According to the regulation ability of distributed resources and the frequency regulation demand, a corresponding dynamic incentive mechanism is constructed:

[0143]

[0144] where is the unit incentive price of the frequency regulation power of the homogeneous cluster in renewable energy; is the unit incentive price of the frequency regulation power of the homogeneous cluster in distributed energy storage; is the unit incentive price of the frequency regulation power of the homogeneous cluster in temperature-controlled loads; h RES,d is a piecewise linear function describing the relationship between the regulation amount of renewable energy and the incentive price; h ESS,d is a piecewise linear function describing the relationship between the regulation amount of distributed energy storage and the incentive price; h TCL,d is a piecewise linear function describing the relationship between the regulation amount of temperature-controlled loads and the incentive price; N RES(t) is the number of renewable energy clusters participating in the t period; N ESS (t) is the number of distributed energy storage clusters participating in the t period; N TCL (t) is the number of temperature control load clusters participating in the t period; is the linear parameter reflecting the regulation amount of the homogeneous cluster A in the u - type distributed resources u,d ; is the linear parameter reflecting the influence of the regulation enthusiasm of the homogeneous cluster A in the u - type distributed resources u,d ; is the proportion of the total incentive obtained by renewable energy in the market clearing price, is the proportion of the total incentive obtained by distributed energy storage in the market clearing price; is the proportion of the total incentive obtained by temperature control load in the market clearing price; f(t) is the system frequency at time t, and f0 is the rated frequency at time t; p FR (T) is the market clearing price of the frequency regulation market at time t;

[0145] Under the dynamic incentive mechanism, the homogeneous cluster A u,d can seek the optimal control strategy to maximize the frequency regulation revenue based on its own state and the leveling incentive price. Therefore, combining the own state of the homogeneous cluster A u,d and the dynamic incentive mechanism, the above - mentioned dynamic stochastic control problem can be defined, that is, to obtain the first optimal control strategy to maximize the frequency regulation revenue:

[0146]

[0147] where is the optimal control strategy at the equilibrium point in the first optimal control strategy;

[0148] Based on the Bellman optimal value principle, the value function v of the above - mentioned first optimal control strategy u,d is defined as:

[0149]

[0150] Furthermore, the HJB equation (Hamilton - Jacobi - Bellman equation) of the above - mentioned first optimal control strategy is:

[0151]

[0152] where is the operator representing the partial derivative with respect to s u,d ; v u,d () is the value function of the optimal control strategy; t is the frequency regulation period; is the standard comprehensive frequency regulation performance index of the homogeneous cluster A u,d ; h u,dIt is a piecewise linear function describing the relationship between the regulation amount of type-u distributed resources and the incentive price; N u (t) is the number of type-u distributed resource clusters participating in the t period; is the frequency regulation power of a single cluster among the type-u distributed resources participating in the t period; is the standard frequency regulation power of a single cluster among the type-u distributed resources participating in the t period; is the penalty coefficient for deviating from the standard frequency regulation power; μ u is the penalty coefficient for deviating from the standard charge; E u,d (t) is the charge of the homogeneous cluster A in the t period u,d ; is the homogeneous cluster A in the t period u,d ;

[0153] It should be noted that to find the optimal strategy point of the first optimal control strategy, all DER clusters need to interact pairwise. Therefore, as the number of clusters increases, searching for the equilibrium point becomes an NP-hard problem. To overcome this problem, the present invention constructs a multi-average field control framework model for approximate solution. In the multi-average field control framework model, DER clusters no longer interact pairwise, but rather the interaction between a single participant in each DER type and the average field formed by the remaining N - 1 participants. When large-scale clusters are involved, the influence of a single cluster's decision on other clusters becomes negligible. However, the influence of the average field on each cluster's decision is significant and can be simulated as a collective effect.

[0154] Based on the multi-average field control framework model, the first optimal control strategy is transformed to obtain the second optimal control strategy, and the HJB equation of the multi-average field control framework model is obtained based on the second optimal control strategy.

[0155] More specifically, the specific process of transforming the first optimal control strategy based on the multi-average field control framework model to obtain the second optimal control strategy includes:

[0156] According to the average field term and the dynamic incentive mechanism, obtain the relationship between the incentive price and the corresponding average field term in the homogeneous cluster:

[0157]

[0158] where is the unit incentive price of the frequency regulation power of the homogeneous cluster in renewable energy; is the unit incentive price of the frequency regulation power of the homogeneous cluster in distributed energy storage; is the unit incentive price of the frequency regulation power of the homogeneous cluster in the thermostatic load; y RES (t) is the average field term describing the collective effect of RES; y ESS(t) is the mean - field term describing the collective effect of ESS; y TCL (t) is the mean - field term describing the collective effect of TCL; h RES,d is a piece - wise linear function describing the relationship between the regulation amount of renewable energy and the incentive price; h ESS,d is a piece - wise linear function describing the relationship between the regulation amount of distributed energy storage and the incentive price; h TCL,d is a piece - wise linear function describing the relationship between the regulation amount of temperature - controlled load and the incentive price; N RES (t) is the number of renewable energy clusters participating in the t - th period; N ESS (t) is the number of distributed energy storage clusters participating in the t - th period; N TCL (t) is the number of temperature - controlled load clusters participating in the t - th period;

[0159] Among them, the mean - field term described in the above steps describes the mapping from the control strategy of a single homogeneous particle to the mean - field effect, and the mean - field term includes:

[0160]

[0161] Among them, y RES (t) is the mean - field term describing the collective effect of RES; y ESS (t) is the mean - field term describing the collective effect of ESS; y TCL (t) is the mean - field term describing the collective effect of TCL; N RES (t) is the number of renewable energy clusters participating in the t - th period; N ESS (t) is the number of distributed energy storage clusters participating in the t - th period; N TCL (t) is the number of temperature - controlled load clusters participating in the t - th period; is the frequency - modulation power of a single cluster in the renewable energy participating in the t - th period; is the frequency - modulation power of a single cluster in the distributed energy storage participating in the t - th period; is the frequency - modulation power of a single cluster in the temperature - controlled load participating in the t - th period.

[0162] Subsequently, optimize the revenue function of the homogeneous cluster A based on the relationship between the incentive price and the mean - field term in the homogeneous cluster u,d :

[0163]

[0164] Among them, J u,d () is the revenue function; s u,d is the decision vector of the u - type distributed resource; y u is the mean - field term describing the collective effect of the u - type distributed resource; is the standard comprehensive frequency - modulation performance index of the homogeneous cluster A u,d ; h u,dis a piecewise linear function describing the relationship between the regulation amount of type-u distributed resources and the incentive price; N u (t) is the number of type-u distributed resource clusters participating in the t time period; y u (t) is the mean field term describing the collective effect of type-u distributed resources in the t time period; is the frequency regulation power of a single cluster among the type-u distributed resources participating in the t time period; is the standard frequency regulation power of a single cluster among the type-u distributed resources participating in the t time period; is the penalty coefficient for deviating from the standard frequency regulation power; μ u is the penalty coefficient for deviating from the standard charge; E u,d (t) is the charge of homogeneous cluster A in the t time period u,d ; is the standard charge of homogeneous cluster A in the t time period u,d ;

[0165] The first optimal control strategy is optimized based on the multi-mean field control framework model to obtain the second optimal control strategy:

[0166]

[0167] The value function of the second optimal control strategy is obtained based on the Bellman optimal value principle:

[0168]

[0169] Based on the value function, the HJB equation of the second optimal control strategy under the multi-mean field control framework model is further obtained:

[0170]

[0171] where is the operator representing the partial derivative with respect to s u,d ; v u,d () is the value function of the optimal control strategy; t is the frequency regulation time period; is the standard comprehensive frequency regulation performance index of homogeneous cluster A u,d ; h u,d is a piecewise linear function describing the relationship between the regulation amount of type-u distributed resources and the incentive price; N u (t) is the number of type-u distributed resource clusters participating in the t time period; is the frequency regulation power of a single cluster among the type-u distributed resources participating in the t time period; is the standard frequency regulation power of a single cluster among the type-u distributed resources participating in the t time period; is the penalty coefficient for deviating from the standard frequency regulation power; μ u is the penalty coefficient for deviating from the standard charge; E u,d (t) is the homogeneous cluster A in the t time periodu,d The charge amount; For the homogeneous cluster A in the t time period u,d The standard charge amount.

[0172] Step 4: Solve the equilibrium solutions of the mean-field term and the HJB equation based on the distributed acceleration method.

[0173] Specifically, as Figure 3 shown, the specific method for solving the equilibrium solutions of the mean-field term and the HJB equation in the above steps is:

[0174] First, select a solution algorithm and set the maximum number of iterations and the convergence parameter of the solution algorithm; in this embodiment, the Mann iterative algorithm is preferably used. The Mann iterative algorithm is a variational inclusion group (VICG) algorithm, mainly used to solve optimization problems in Banach spaces. The basic idea of this algorithm is to update the current iteration point by solving a variational inclusion group (VICG) in each iteration, and it has good convergence performance.

[0175] Combine each homogeneous cluster A u,d with the HJB equation of the multi-mean-field control framework model to obtain the frequency modulation power corresponding to the homogeneous cluster A in the k-th iteration u,d respectively;

[0176] And calculate the mean-field term corresponding to the homogeneous cluster A in the k-th iteration according to the frequency modulation power corresponding to the homogeneous cluster A in the k-th iteration u,d respectively; u,d The mean-field term;

[0177] When the mean-field term result corresponding to the homogeneous cluster A in the k-th iteration u,d meets the convergence parameter, then output the frequency modulation power corresponding to the homogeneous cluster A in the k-th iteration u,d respectively; otherwise, increment the current iteration number by 1 and calculate the frequency modulation power and the mean-field term result corresponding to the homogeneous cluster A in the (k + 1)-th iteration in turn u,d respectively.

[0178] Optionally: The distributed acceleration algorithm uses the Mann iterative algorithm:

[0179]

[0180] Wherein, is the mean-field term corresponding to the homogeneous cluster A in the k-th iteration u,d respectively; is the mean-field term corresponding to the homogeneous cluster A in the (k + 1)-th iteration u,d respectively; is the coefficient sequence of the Mann iterative algorithm at the (k + 1)-th iteration; N uis the number of DER clusters of type u; is the homogeneous cluster A corresponding to the (k + 1)-th iteration u,d of the frequency modulation power.

[0181] Furthermore, in this embodiment, to verify the effect of the above method, a simulation is carried out. The device for the simulation is an HPZ600 machine, which is configured with an Intel Xeon CPU (frequency 2.4 GHz) and 12 GB of RAM. The above device runs the simulation model in the MATLABR2023a environment by calling the Gurobi 9.5.2 solver, and the time step is set to 5 min.

[0182] First, taking N = 100 as an example, the optimal frequency modulation control strategy of the DER cluster at the MFE point is as Figure 4 and Figure 5 shown: The growth trends of the incentive price and the frequency modulation power are the same, which indicates that the multi-average field control method proposed by the present invention has incentive compatibility and can well motivate the DER cluster to participate in frequency modulation. When the incentive price rises, the DER cluster will actively respond to the frequency modulation instruction of the VPP, and the incentive price will also increase with the improvement of the frequency modulation performance until it reaches the upper limit. In the time dimension, the change trends of the frequency modulation power and the incentive price of different clusters are similar. The above results show that the dynamic granulation method adopted by the present invention can effectively divide the DERs into homogeneous clusters with similar external regulation characteristics.

[0183] Secondly, the net benefits of different DER clusters are as Figure 6 shown: The net benefit of the cluster gradually increases with the increase of d and reaches a peak value of 4427.74 when d = 72. At this time, the incentive sensitivity of the cluster to the frequency down-regulation demand is 0.034, and the incentive sensitivity of the cluster to the frequency up-regulation demand is 0.052. This is because the frequency modulation power and the corresponding incentive price increase with the increase of d, and the net income always increases. When d ∈ [73, 100], the net income will drop sharply because the income and frequency modulation power that can be allocated by the VPP are limited. If the total output of the DER cluster exceeds the frequency modulation power, it will be economically penalized. This shows that the penalty factor can effectively limit the response frequency modulation power of the DER cluster so that it does not exceed the winning bid frequency modulation power of the VPP.

[0184] Finally, the average computation times of the multi-average field control method of the present invention and the traditional dynamic stochastic control under different numbers of participating DER clusters are shown in Table I. When N = 100, the computation time of the multi-average field control method is 0.129% and 0.017% shorter than that of the dynamic stochastic control. As N increases, the computation time of the dynamic stochastic control increases sharply and exceeds 24 hours when N = 700, making it no longer applicable to the revenue distribution in the frequency regulation market. The computation time of the multi-average field control increases slowly with the increase of the cluster number N because the proposed distributed acceleration algorithm based on granular computing allows parallel computing of clusters and speeds up the convergence rate.

[0185] Table I Comparison of Computation Times of Different Game Modes

[0186]

[0187] In addition, the present invention also discloses a multi-average field control device for a virtual power plant to participate in secondary frequency regulation. The device includes:

[0188] A modeling and evaluation module for evaluating the adjustable space and comprehensive frequency regulation performance of heterogeneous distributed resources inside the virtual power plant by using a unified modeling method;

[0189] A cluster division module for respectively dividing different types of distributed resources into multiple homogeneous clusters with the same response characteristics by using a dynamic granulation method;

[0190] A dynamic analysis module for constructing a multi-average field control framework model and describing the dynamic interaction between a single homogeneous cluster and the average field term in the same type of distributed resources through the average field term and the HJB equation;

[0191] A result solving module for solving the equilibrium solutions of the average field term and the HJB equation based on a distributed acceleration method.

[0192] Furthermore, the present invention also discloses a non-volatile storage medium. The non-volatile storage medium includes a stored program. When the program runs, it controls the device where the non-volatile storage medium is located to execute the above multi-average field control method.

[0193] In addition, the present invention also discloses a device including a processor and a memory; computer-readable instructions are stored in the memory, and the processor is used to run the computer-readable instructions. When the computer-readable instructions run, they execute the above multi-average field control method.

[0194] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable non-transitory storage media (including but not limited to disk memory, CD-ROM, optical memory, etc.) that contain computer-usable program code.

[0195] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatuses, and computer program products according to embodiments of the present invention. It should be understood that each flow and / or block in the flowchart and / or block diagram, and the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices produce means for implementing the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0196] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory produce a manufactured article including instruction means that implement the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0197] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0198] The above is only the preferred embodiment of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the technical principle of the present invention, several improvements and modifications can be made, and these improvements and modifications should also be regarded as the protection scope of the present invention.

Claims

1. A multi-average field control method for virtual power plants participating in secondary frequency modulation, characterized in that: The following steps are involved: A unified modeling approach is used to evaluate the adjustable space and comprehensive frequency regulation performance of heterogeneous distributed resources within a virtual power plant. A dynamic granulation method is used to divide different types of distributed resources into multiple homogeneous clusters with the same response characteristics; A multi-mean field control framework model is constructed, and the dynamic interaction between a single homogeneous cluster and mean field terms in the same distributed resource is described by the mean field terms and HJB equations: Based on the purpose of seeking the optimal frequency modulation instructions and economic benefit distribution, a dynamic random control model is constructed according to the interaction between distributed resource clusters; Obtaining homogeneous cluster A based on dynamic random control model u,d The profit function is: According to the regulation capability and frequency regulation requirements of distributed resources, a corresponding dynamic incentive mechanism is constructed: Combining the state of the homogeneous cluster and the dynamic incentive mechanism, the first optimal control strategy to maximize the frequency modulation benefits is obtained: Based on the multi-mean field control framework model, the first optimal control strategy is transformed to obtain the second optimal control strategy, and the HJB equation of the multi-mean field control framework model is obtained based on the second optimal control strategy: According to the average field term and dynamic incentive mechanism, the relationship between the incentive price and the corresponding average field term in the homogeneous cluster is obtained: Optimizing homogeneous cluster A based on the relationship between incentive price and mean field term in homogeneous clusters u,d The profit function is: Based on the multi-mean field control framework model, the first optimal control strategy is optimized to obtain the second optimal control strategy: Based on Bellman's optimal value principle, the value function of the second optimal control strategy is obtained: Based on the value function, the HJB equation of the second optimal control strategy under the multi-mean field control framework model is obtained: The mean field term and the equilibrium solution of the HJB equation are solved based on the distributed acceleration method; in, is the optimal control strategy at the equilibrium point in the first optimal control strategy; RES is renewable energy, ESS is distributed energy storage, and TCL is temperature control load; J u，d ( ) is the profit function; s u，d is the decision vector of u-type distributed resources; S u，-d is the joint decision vector of u-type distributed resource clusters after excluding homogeneous clusters; It is the standard comprehensive FM performance index for homogeneous clusters; is the unit incentive price of frequency modulation power of homogeneous clusters in u-type distributed resources; is the frequency modulation power of a single cluster of u-type distributed resources participating in period t, is the corresponding standard FM power; μ u is the penalty coefficient for deviation from the standard charge; E u，d (t) is the charge of the homogeneous cluster during period t, is the corresponding standard charge; and The deviation from the standard FM power is The penalty coefficient h u，d is a piecewise linear function describing the relationship between the regulation amount and incentive price of u-type distributed resources; N u (t) is the number of u-type distributed resource clusters participating in time period t; is a linear parameter reflecting the amount of homogeneous cluster regulation in u-type distributed resources, is the linear parameter reflecting the positive influence of homogeneous cluster regulation in u-type distributed resources; is the ratio of the total incentives obtained by u-type distributed resources to the market clearing price; f(t) is the system frequency in period t, f0 is the rated frequency in period t; p FR (t) is the FM market clearing price in period t; y u (t) is the average field term describing the collective effect of u-type distributed resources in period t; Indicates S u,d Operator for partial differentiation; v u，d ( ) is the value function of the optimal control strategy.

2. The multi-mean field control method according to claim 1, characterized in that: The dynamic random control model is: in is the set of participants, which represents the number of u-type distributed resource clusters participating in period t, and is the state space, which represents the joint state of all clusters, and is the decision vector space, which represents the joint control strategy of all clusters, and is the utility function, which represents the degree of preference of the participants in the interaction.

3. The multi-mean field control method according to claim 1, characterized in that: The steps of evaluating the adjustable space and comprehensive frequency regulation performance of heterogeneous distributed resources within the virtual power plant by using a unified modeling method include: According to the operating characteristics of heterogeneous distributed resources, a convex set in T-dimensional space is used Unified modeling of the aggregate power feasible domain of distributed resources: Where u is the type of distributed resource, which includes renewable energy, distributed energy storage or temperature control load; F u is a coefficient matrix, and Where K represents the number of constraints in a time interval; b u is a constant matrix, and Where T represents the number of scheduling time intervals; is the power of distributed resource i of type u, is the charge vector of u-type distributed resource i, where Using the Minkowski sum, Mapping to the aggregate power feasible region Where n is the number of distributed resources; is the sum of the points in Euclidean space; is the aggregate power of u-type distributed resources; The embedded rectangular pyramid method is used to solve the operating power boundaries of different time periods to achieve time decoupling of aggregated adjustable power: in and are the upper and lower bounds of the operating power; is the adjustable power range of u-type distributed resources; Finally, the three indicators of regulation rate, regulation accuracy and response time are selected to form the comprehensive indicators for the performance evaluation of distributed resources participating in frequency regulation: in, is a comprehensive indicator; To adjust the rate; To adjust the accuracy; is the response time; K i,1 is the adjustment rate of a single distributed resource i; K i,2 is the adjustment accuracy of a single distributed resource i; K i,3 is the response time of a single distributed resource i.

4. The multi-mean field control method according to claim 1, characterized in that: The steps of using the dynamic granulation method to divide different types of distributed resources into multiple homogeneous clusters with the same response characteristics include: First, a four-dimensional information system containing detailed information on distributed resources is constructed: S = (U, D, V, f); Where U = {DER u,1 ,DER u,2 ,…,DER u,n } is a domain, and DER u,i (i=1,2,…,n) represents the i-th distributed resource in the u-type distributed resource set; D is the attribute set; V is the attribute value set; f is the information function, which represents the mapping between each distributed resource and the corresponding attribute value; Randomly aggregate the distributed resources in the domain U into cluster A u,c , where cluster A u,c The conditional attributes are: in For cluster A u,c The aggregate power can be adjusted downwards within the range; For cluster A u,c The aggregate power can be adjusted upwards within the range; For cluster A u,c Comprehensive frequency modulation performance index; D m For attributes; According to the time decoupling method of aggregate adjustable power and the comprehensive index calculation method, cluster A is obtained. u,c The value set of the attribute According to cluster A u,c The attribute value set V and the corresponding information function to obtain the value of each attribute are: in, D m The value of For cluster A u,c The upper limit of aggregate power; For cluster A u,c The lower limit of aggregate power; For cluster A u,c Base value of aggregate power; K i,1 is the adjustment rate of a single distributed resource i; K i,2 is the adjustment accuracy of a single distributed resource i; K i,3 is the response time of a single distributed resource i; Determine cluster A u,c and standard cluster A u,s Is there an indistinguishable relationship between where Euc[f(A u,c ,D),f(A u,s ,D)] is cluster A u,c and standard cluster A u,s The Euclidean distance between the standard cluster A u,s The constraints are as follows, will be with standard cluster A u,s Clusters A with indistinguishable relationships between u,c As a homogeneous cluster A u,d : Where E u,d A is a homogeneous cluster u,d The charge of A is a homogeneous cluster u,d The power increase; A is a homogeneous cluster u,d The power reduction of Standard cluster A u,s Increase the power limit; Standard cluster A u,s Lower the power limit; Standard cluster A u,s The upper limit of the charge; Standard cluster A u,s The lower limit of charge; A is a homogeneous cluster u,d Comprehensive frequency modulation performance indicators; A is a homogeneous cluster u,d Standard comprehensive frequency modulation performance indicators.

5. The multi-mean field control method according to claim 1, characterized in that: The mean field term includes:

6. The multi-mean field control method according to claim 1, characterized in that: The step of solving the mean field term and the equilibrium solution of the HJB equation comprises: Select a solution algorithm and set the maximum number of iterations and convergence parameters of the solution algorithm; Each homogeneous cluster A u,d Combined with the HJB equation of the multi-mean field control framework model, the homogeneous cluster A corresponding to the kth iteration is obtained u,d FM power; And according to the homogeneous cluster A corresponding to the kth iteration u,d The frequency modulation power calculation obtains the homogeneous cluster A corresponding to the kth iteration u,d The mean field term of When the kth iteration corresponds to the homogeneous cluster A u,d When the average field term result satisfies the convergence parameter, the homogeneous cluster A corresponding to the kth iteration is output u,d Otherwise, the current iteration number is incremented by 1 and the homogeneous cluster A corresponding to the k+1th iteration is calculated in sequence u,d The FM power and mean field term results.

7. The multi-mean field control method according to claim 5, characterized in that: The distributed acceleration method adopts the Mann iterative algorithm: in, The kth iteration corresponds to the homogeneous cluster A u,d The mean field term of The k+1th iteration corresponds to the homogeneous cluster A u,d The mean field term of is the coefficient sequence of the Mann iterative algorithm at the k+1th iteration; N u is the number of DER clusters of type u; The k+1th iteration corresponds to the homogeneous cluster A u,d FM power.

8. A multi-mean field control device for a virtual power plant participating in secondary frequency modulation, using the multi-mean field control method according to any one of claims 1 to 7, characterized in that: The device comprises: A modeling and evaluation module is used to evaluate the adjustable space and comprehensive frequency regulation performance of heterogeneous distributed resources within the virtual power plant using a unified modeling method; The cluster partitioning module is used to divide different types of distributed resources into multiple homogeneous clusters with the same response characteristics using a dynamic granulation method; Dynamic analysis module, which is used to construct a multi-mean field control framework model and describe the dynamic interaction between a single homogeneous cluster and mean field terms in the same distributed resource through mean field terms and HJB equations; The result solving module is used to solve the equilibrium solution of the multi-mean field control framework model and the HJB equation based on the distributed acceleration method.

Images