Active power distribution network primary frequency modulation optimization coordination control method under weak model condition

The three-layer control architecture and linear optimization model constructed by the Koopman operator solves the nonlinear problems and parameter dependence problems in the primary frequency modulation optimization of the distribution network, realizes frequency characteristic optimization and trend safety constraints, and improves the calculation speed and accuracy.

CN119029937APending Publication Date: 2024-11-26TIANJIN UNIV
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202411163940.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-23
Publication Date
2024-11-26

AI Technical Summary

Technical Problem

The existing distribution network's primary frequency modulation optimization method is difficult to meet the online frequency modulation requirements, the traditional physical analysis model is highly nonlinear and difficult to analyze, the model output of the data-driven method is weak, the linear data-driven flow model has low accuracy, and it is unable to adapt to the new high permeability. Strong nonlinear scenarios for energy access.

Method used

Using the three-layer control architecture of Koopman operator, the current model is constructed and the sag coefficient and reactive power adjustment of DER are optimized. The timing coupled algebraic equation of the system frequency response is obtained through differential discretization, so as to achieve algebraic and minimize network loss and voltage deviation. , establish a linear optimization model based on Koopman global sensitivity, and transform non-convex non-linear optimization into easy-to-solve linear optimization.

Benefits of technology

The frequency characteristics optimization of distribution network, reasonable power distribution and fast frequency response of distributed power supplies are realized, the frequency modulation cost and voltage deviation are reduced, the calculation speed is improved, the dependence on model parameters is avoided, and the current safety constraints are met.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119029937A_ABST
    Figure CN119029937A_ABST
Patent Text Reader

Abstract

The invention relates to an active power distribution network primary frequency modulation optimization coordination control method under a weak model condition. The method comprises the following steps: (1) constructing a power flow model of Koopman state space dimension raising; (2) an active power distribution network primary frequency modulation optimization control model is constructed, the optimization target is to optimize the droop coefficient and the reactive power regulation quantity of DER in multiple time periods, and the algebraic sum of network loss and voltage deviation is minimum on the premise that the overall frequency modulation characteristic, the power flow constraint and the like are met; and (3) solving the established primary frequency modulation optimization control model, and realizing active power distribution network primary frequency modulation optimization coordination control considering the power flow safety constraint under a weak model condition.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of primary frequency regulation of distribution networks, and in particular to an active distribution network primary frequency regulation optimization and coordination control method taking into account power flow safety constraints under weak model conditions. Background Art

[0002] The participation of controllable distributed power generation clusters in the active distribution network in frequency regulation services can effectively improve the frequency safety and stability of the power system. In order to meet the rapidity requirements of primary frequency regulation, the widely used method to achieve the participation of distributed power generation in frequency regulation control is to attach a controller to the converter so that it has a fast inertia response capability. Reference [1] uses virtual synchronous generator technology to make the photovoltaic grid-connected inverter equivalent to a traditional synchronous generator, thereby participating in grid frequency regulation. By simulating the power-frequency static characteristics of conventional synchronous generators, reference [2] uses a droop control strategy to control wind turbines to participate in grid frequency regulation services and adjust the droop coefficient under different wind speeds. Under the above control method, although distributed power generation can participate in the primary frequency regulation of the power system, it lacks cluster-based coordination and cooperation, and it is difficult to accurately respond to the overall cluster frequency regulation instructions issued by the grid operator during the frequency regulation process.

[0003] In order to achieve the coordination and unification of distributed renewable energy power distribution and the overall frequency regulation characteristics of the cluster, reference [3] aggregates different types of distributed power sources into a virtual power plant based on the synchronous virtual power plant framework of distributed energy with grid-connected inverter interface to provide auxiliary frequency regulation services for the power grid. Reference [4] designs the droop coefficient of each distributed power source in the distribution network to provide frequency regulation services for the power grid at the head end of the feeder, and uses the participation factor derived from the linearized power flow to evaluate the contribution of each distributed power source to the overall frequency regulation. References [5] and [6] respectively study the primary frequency regulation service provided by electric vehicles from the perspectives of stability and economy. Based on the control of delivered power, a charging and discharging power algorithm for electric vehicles to participate in power grid frequency regulation is formulated. However, the above methods do not consider the power flow constraints within the distribution network, which may lead to problems such as node voltage exceeding the limit and line power overload, and pose safety hazards. Based on the optimal power flow model, reference [7] constructs an active distribution network primary frequency regulation optimization model taking into account power flow constraints. By coordinating and optimizing the droop coefficient and reactive power of each distributed power source, the optimal power flow distribution is achieved while ensuring the overall frequency regulation characteristics. However, this method only considers three sections of system frequency change, which cannot fully reflect the whole process of primary frequency regulation. At the same time, the distribution network frequency regulation optimization model established by this method belongs to a non-convex non-deterministic polynomial problem, which is difficult to solve. Solving based on existing heuristic algorithms is relatively easy to implement, but the calculation time is long and it is easy to fall into local optimality. [8-10] The computational complexity of the solution is low based on analytical methods such as second-order cone programming relaxation. [11-12]However, the accuracy of relaxation needs to meet certain hard conditions. When the objective function is not just a single objective network loss, the relaxation will be inaccurate. In addition, since the above two solution methods are heavily dependent on model parameters, and the real-time topology information and precise line parameters of the medium and low voltage distribution network are usually difficult to obtain accurately

[13] , so high accuracy cannot be guaranteed in practical applications.

[0004] The widespread deployment of high-precision measurement terminals has promoted the development and application of data-driven methods in the field of distribution network control research, and provided an effective solution to overcome the problems of inaccurate parameters, time-consuming iterative solutions, and lack of convergence in traditional physical model-based distribution network optimization control methods. Reference

[14] uses deep reinforcement learning methods to optimize the day-ahead dispatching strategy of active distribution networks to ensure safe, stable and efficient operation of the system. Reference

[15] constructs a convolutional neural network model and uses historical data to train the correspondence between load patterns and their optimal topology, thereby improving the online calculation speed of distribution network reconstruction. However, the optimization model established by the above method is a black box, completely separated from the physical model, and the interpretability of the model output results is weak. In some cases, it is difficult to ensure the rationality of the output results and cause impacts on the operation of the distribution network. Different from the above methods, references [16-17] combine model linearization and data-driven methods based on a data-mechanism hybrid method, obtain power flow sensitivity information by training the historical operation data of the distribution network, and then convert the nonlinear optimization model into a linear optimization model. Similarly, the optimization model is quickly solved without relying on model parameters, and has good stability and interpretability. However, the linear data-driven models (i.e., linear parameter regression) used by these methods can only maintain high accuracy within a small range of power fluctuations and cannot adapt to the strong nonlinear scenarios of high penetration of new energy access. At the same time, they only perform sensitivity modeling for specific operating points and require continuous online data updates to adapt to the current operating status, and are unable to estimate the global sensitivity.

[0005] In summary, the current primary frequency regulation optimization methods of distribution networks still have certain defects and shortcomings. Due to the large number of distributed power sources in the distribution network and the complex dynamic characteristics of the power flow, the optimization model of the traditional physical analysis method is highly nonlinear and difficult to solve analytically, and it is difficult to meet the requirements of online frequency regulation optimization. In addition, it is difficult to obtain the line parameters of the medium and low voltage distribution network, and the calculation accuracy of the physical analysis method cannot be guaranteed. In terms of data-driven methods, although the neural network method has been widely used, it is separated from the physical model, resulting in weak interpretability of the model output results. The linear data-driven power flow model has low accuracy and cannot adapt to the strong nonlinear scenarios brought about by the access of high-penetration new energy.

[0006] Related Literature

[0007] [1]C.Li and S.Yuan,“Research on frequency modulation controlofphotovoltaic power generation system based on VSG,”International Journal ofLow-Carbon Technologies,vol.16,no.2,pp.287–293,May.2021.

[0008] [2]K.V.Vidyanandan and N.Senroy,“Primary frequency regulation bydeloaded wind turbines using variabledroop,”IEEE Trans.Power Syst.,vol.28,no.2,pp.837–846,May.2013.

[0009] [3]Q.Hu and Z.Wu,“Grid-forming inverter enabled virtual power plantswith inertia support capability,”IEEE Trans.Smart Grid,vol.13,no.5,pp.4134–4143,Sept.2022.

[0010] [4]S.S.Guggilam,C.Zhao,E.Dall’Anese,Y.C.Chen,and S.V.Dhople,“Optimizing power–frequency droop characteristics of distributed energyresources,”IEEE Trans.Power Syst.,vol.33,no.3,pp.3076–3086,May.2018.

[0011] [5]Z.Yang,F.Yang,W.Hu,Z.Zhang,and X.Zhou,“Delay compensation controlstrategy for electric vehicle participating in frequency regulation based onMPC algorithm,”Electronics,vol.11,no.15,pp.2341,Jul.2022.

[0012] [6]J.Donadee,“Stochastic optimization of grid to vehicle frequencyregulation capacity bids,”IEEE Trans.Smart Grid,vol.5,no.2,pp.1061–1069,March.2014.

[0013] [7]E.O.Kontis, R.d.Nozal,J.M.Mauricio,and C.S.Demoulias,“Provisionofprimary frequency response as ancillary service from active distributionnetworks to the transmission system,”IEEE Trans.Smart Grid,vol.12,no.6,pp.4971–4982,Nov.2021.

[0014] [8]R.Pegado,Z. Y.Molina,and C.Castillo,“Radial distributionnetwork reconfiguration for power losses reduction based on improvedselective BPSO,”Electric Power Systems Research,vol.169,pp.206–213,Apr.2019.

[0015] [9]H.Ikegami and H.Mori,“Development ofDEPSO Island model withparticle speed limit for distribution network reconfigurations,”IFAC-PapersOnLine,vol.51,no.28,pp.552–557,Dec.2018.

[0016]

[10] T.T.Nguyen,B.H.Dinh,N.V.Quynh,M.Q.Duong,and L.V.Dai,“Anovelalgorithm for optimal operation ofhydrothermal power systems underconsidering the constraints in transmission networks,”Energies,vol.11,2018.

[0017]

[11] F.Sun,M.Yu,Q.Wu,and W.Wei,“A multi-time scale energy managementmethod for active distribution networks with multiple terminal soft openpoint,”International Journal ofElectrical Power&Energy Systems,vol.128,2021.

[0018]

[12] T.Ding,C.Li,Y.Yang,J.Jiang,Z.Bie,and F.Blaabjerg,“Atwo-stagerobust optimization for centralized-optimal dispatch of photovoltaicinverters in active distribution networks,”IEEE Trans.Sustain.Energy,vol.8,no.2,pp.744–754,Apr.2017.

[0019]

[13] Y.Wang,Q.Chen,T.Hong,and C.Kang,“Review of smart meter dataanalytics:applications,methodologies,and challenges,”IEEE Trans.Smart Grid,vol.10,no.3,pp.3125–3148,May.2019.

[0020]

[14] X.Li,X.Han,and M.Yang,“Day-ahead optimal dispatch strategy foractive distribution network based on improved deep reinforcement learning,”IEEE Access,vol.10,pp.9357–9370,2022.

[0021]

[15] Y.Yu,M.Yang,Y.Zhang,P.Ye,X.Ji,and J.Li,“Fast reconfigurationmethod of low-carbon distribution network based on convolutional neuralnetwork,”Front.Energy Res.,vol.11,p.1102949,Jan.2023.

[0022]

[16] H.Su,P.Li,P.Li,X.Fu,L.Yu,and C.Wang,“Augmented sensitivityestimation-based voltage control strategy of active distribution networkswith PMU measurement,”IEEEAccess,vol.7,pp.44987–44997,2019.

[0023]

[17] H. Xu, AD Dominguez-Garcia, and PWSauer, “A data-driven voltage control framework for power distribution systems,” IEEE Power & Energy Society General Meeting, Portland, OR, USA, Aug. 2018, pp. 1–5. Summary of the invention

[0024] In view of the above problems, the present invention proposes an optimization and coordinated control method for primary frequency regulation of active distribution network based on Koopman operator and taking into account power flow safety constraints, and constructs a three-layer control architecture for distribution network participating in primary frequency regulation.

[0025] A method for optimizing and coordinating primary frequency regulation of an active distribution network under weak model conditions comprises the following steps:

[0026] (1) Construct a power flow model with Koopman state space dimension increase;

[0027] (2) Construct an active distribution network primary frequency regulation optimization control model. Its optimization goal is to optimize the droop coefficient and reactive power regulation of DER in multiple time periods, aiming to minimize the algebraic sum of network loss and voltage deviation under the premise of satisfying the overall frequency regulation characteristics and power flow constraints. Its constraints are: the time-series coupled algebraic equation of the system frequency response is obtained through differential discretization. When the frequency deviation exceeds the dead zone, each distributed power source in the distribution network provides active power support through its own droop characteristics. The voltage amplitude of each node and the current amplitude of each line are calculated to determine whether the power flow inside the distribution network is within a safe and reasonable range to avoid voltage and current exceeding the limit. The power of each photovoltaic unit needs to meet the capacity limit, and each energy storage unit needs to meet the power and power capacity constraints.

[0028] (3) Solve the established primary frequency regulation optimization control model to realize the primary frequency regulation optimization coordinated control of the active distribution network taking into account the power flow safety constraints under weak model conditions.

[0029] Furthermore, the method of step (1) is as follows:

[0030] The basic form of the power flow equation is as follows:

[0031]

[0032] Among them, the input variable is x = [PQ] T , which is composed of active power and reactive power injected by the node, and the output variable y = [V IP pcc ] TIt is composed of node voltage, line current, and active power at the grid connection point. C represents the linearized power flow mapping matrix, x lift represents the high-dimensional input variable after the state space is upgraded, and ψ(x) represents the dimensionality-elevating operation function of the input variable x;

[0033] When the dimension is increased to n, ψ(x) is composed of n scalar functions ψ i (x) constitutes, namely:

[0034] ψ i (x) = f lift (xc i ) (2)

[0035]

[0036] Among them, f lift represents the expanded form of the dimension-raising function, c i represents the expanded i-th dimension basis vector, whose value is selected as a random value within the range of the distribution network input variable, c ij Represents c i The jth element in , K is the dimension of the input variable x;

[0037] When there are S time-section historical data as training sample sets, the input variable sample set is The output variable sample set is Based on the historical operation data of the distribution network, data-driven training is carried out, and the linearized power flow mapping matrix C can be obtained through least squares estimation. The calculation formula is as follows:

[0038]

[0039] in, represents the Moore-Penrose inverse of a matrix;

[0040] make They represent the voltage amplitude of node i, the current amplitude of the lth line (l∈E, indicating the branch number in the corresponding branch set), and the sensitivity of the active power of the grid connection point to the active power injected into node j; the linearized power flow mapping matrix C driven by Koopman data, the sensitivity The calculation formulas are as follows:

[0041]

[0042]

[0043]

[0044] Among them, V i ,Il , P PCC They are respectively represented as node voltage, line current and active power at the grid connection point, N is the total number of distribution network nodes, C ij , C (N+l),j , C 2N,j Represent the corresponding output variables V in the linearized power flow mapping matrix C i ,I l , P PCC With input variable p j The element C i,(K+k) , C (N+l),(K+k) , C 2N,(K+k) Represent the corresponding output variables V in the linearized power flow mapping matrix C i ,I l , P PCC and the kth dimension-raising variable ψ k (x) is an element; the partial differential term on the right side of the equations (5)-(7) can be calculated as follows:

[0045]

[0046] According to the above calculation process, the node active injection power in the denominator of equations (5)-(8) is replaced by the node reactive power, and the above three state variables V are derived in the same way. i ,I l , P PCC Sensitivity matrix with reactive power injected into the node.

[0047] Furthermore, in step (2), the optimization objective function established is:

[0048] The optimization objective function can be expressed as:

[0049]

[0050]

[0051] ΔQ=(ΔQ 1 ,…,ΔQ t ,…,ΔQ T ) (11)

[0052]

[0053]

[0054]

[0055] Among them, the first term of the objective function represents the network loss, and the second term represents the voltage distribution; C1, C2 are weight coefficients, t is the time point in the optimization period, T is the number of time points in the optimization period, N is the total number of distribution network nodes, B represents the total number of optimized particles of DER reactive regulation in the optimization period, the symbol < > represents rounding up, n and m represent the number of photovoltaic and energy storage included in the distribution network, respectively, Γ represents the set of nodes connected to photovoltaic, Π represents the set of nodes connected to energy storage, Υ represents the set of nodes connected to the load, represents the initial active power of the photovoltaic at node i, represents the initial active power of energy storage at node i, represents the load active power at node i, They represent the active regulation of photovoltaic power and energy storage at node i at time t respectively; represents the active power of the grid connection point at time t, represents the node voltage, K f represents the (m+n)×1-order column vector consisting of the droop coefficients of the independent variable DER, ΔQ t and The (m+n)×1-order column vectors correspond to the reactive power regulation of DER at the t-th time and the k-th optimization granularity. ΔQ represents the reactive power dispatch sequence of each DER in the optimization period, which is composed of T ΔQ t A (m+n)×T matrix.

[0056] Furthermore, the constraint conditions in step (2) are specifically:

[0057] The time-series coupled algebraic equations of the system frequency response obtained by differential discretization are as follows:

[0058]

[0059] Among them, t1 represents the frequency discrete point, Δt1 represents the frequency discrete interval, f represents the system real-time frequency, H represents the power system inertia coefficient, P0 represents the initial active output of the distribution network, P G Represents the active output of the conventional electric field, P L Indicates the active load of the power grid, K G and K f They represent the droop coefficients of conventional power stations and distribution networks respectively;

[0060] When the frequency deviation exceeds the dead zone, each distributed power source in the distribution network provides active power support through its own droop characteristics, as shown in the following formula:

[0061]

[0062]

[0063] in, and Represent the active power of photovoltaic and energy storage respectively, and Represent the active initial power of photovoltaic and energy storage respectively, and are the droop coefficients of PV and energy storage, respectively;

[0064] Calculate the voltage amplitude of each node and the current amplitude of each line to determine whether the internal power flow of the distribution network is within a safe and reasonable range to avoid voltage and current exceeding the limit, as shown in the following formula:

[0065]

[0066] In the formula, They represent the vectors composed of the voltage amplitudes of each node and the line current amplitudes in the initial state respectively; Indicates the active power of the grid connection point in the initial state; X P ∈R N×N , X Q ∈R N×N are the voltage-node active sensitivity mapping matrix and the voltage-node reactive sensitivity mapping matrix, Y P ∈R (N-1)×N , Y Q ∈R (N-1)×N are the current-node active sensitivity mapping matrix and the current-node reactive sensitivity mapping matrix respectively; Z P ∈R 1×N , Z Q ∈R 1×N They are the grid-connected point active power-node active power sensitivity mapping matrix and the grid-connected point-node reactive power sensitivity mapping matrix; ΔP t , ΔQ t They represent the vectors composed of the active and reactive regulation quantities of each active node at time t; R and S represent the node position matrices of photovoltaic and energy storage in the distribution network respectively; They represent the vectors composed of the active changes of photovoltaic power and energy storage at time t respectively; They represent the vectors composed of the reactive power changes of photovoltaic and energy storage at time t respectively;

[0067] The power of each photovoltaic unit needs to meet the capacity limit, which is expressed as follows:

[0068]

[0069] in, represents the maximum active output of the ith photovoltaic unit; S i,pv is the rated capacity of the corresponding ith PV unit;

[0070] Each energy storage unit must meet the power and power capacity constraints, as shown in the following formula:

[0071]

[0072]

[0073]

[0074] in, is the energy storage capacity connected to the i-th node at time t; and are the corresponding energy storage charging power and discharging power respectively; η charge and η discharge are the charge and discharge efficiencies respectively; Δt is the optimization time interval; and They are the upper limits of the charging and discharging power of energy storage respectively; Indicates the charge and discharge mark of energy storage, and introduces Constraint means that at any time, the energy storage can only be in one of the three states: charging, discharging, or neither charging nor discharging. It is the energy storage capacity limit.

[0075] Compared with the prior art, the present invention has the following advantages:

[0076] (1) A three-layer control method is proposed to solve the contradiction between the good frequency characteristics of the distribution network and the reasonable power allocation of distributed generation, fast frequency response and optimized solution time, and get rid of the heavy dependence on model parameters.

[0077] (2) A coordinated optimization model for primary frequency regulation of active distribution networks taking into account power flow safety constraints is constructed. This model fully considers the entire process of system frequency change and synergistically optimizes the DER droop coefficient and reactive power. It can fully tap the ability of clusters to participate in frequency support and minimize frequency regulation costs and voltage deviations.

[0078] (3) A linear optimization solution method for primary frequency regulation of active distribution networks based on Koopman global sensitivity is proposed, which transforms the original difficult-to-solve non-convex nonlinear optimization model into an easy-to-solve linear optimization model, greatly improving the calculation speed. BRIEF DESCRIPTION OF THE DRAWINGS

[0079] Figure 1 This is a flow chart of the method proposed by the present invention for primary frequency regulation control of active distribution network

[0080] Figure 2 This is the topological structure diagram of the 10kV distribution network used in the simulation case of the present invention.

[0081] Figure 3 The method proposed in this invention is the Koopman linear optimization method which optimizes only the active power.

[0082] Comparison of network loss results of the method and particle swarm optimization method in four different scenarios

[0083] Figure 4 The method proposed in this invention is the Koopman linear optimization method which optimizes only the active power.

[0084] Comparison of voltage deviation results of the particle swarm optimization method and the particle swarm optimization method in four different scenarios

[0085] Figure 5 The method proposed in this invention is the Koopman linear optimization method which optimizes only the active power.

[0086] Comparison of optimization results of particle swarm optimization method in four different scenarios

[0087] Figure 6 The method proposed in this invention is the Koopman linear optimization method which optimizes only the active power.

[0088] Comparison of power-frequency characteristic curves of the method and particle swarm optimization method in scenario 1

[0089] Figure 7 The method proposed in this invention is the Koopman linear optimization method which optimizes only the active power.

[0090] Comparison of power-frequency characteristic curves of the method and particle swarm optimization method in scenario 3

[0091] Figure 8 The method proposed in this invention is the Koopman linear optimization method which optimizes only the active power.

[0092] Frequency change curve comparison of method and particle swarm optimization method in scenario 1

[0093] Fig. 9 The method proposed in this invention is the Koopman linear optimization method which optimizes only the active power.

[0094] Frequency change curve comparison of method and particle swarm optimization method in scenario 3

[0095] Fig.10 The method proposed in this invention is the Koopman linear optimization method which optimizes only the active power.

[0096] Comparison of voltage distribution between the method and particle swarm optimization method in scenario 1

[0097] Fig.11 The method proposed in this invention is the Koopman linear optimization method which optimizes only the active power.

[0098] Comparison of voltage distribution between the method and particle swarm optimization method in scenario 4

[0099] Fig.12 The method proposed in this invention is the Koopman linear optimization method which optimizes only the active power.

[0100] Comparison of voltage distribution of the method and particle swarm optimization method at the stable moment of scenario 1

[0101] Fig.13 The method proposed in this invention is the Koopman linear optimization method which optimizes only the active power.

[0102] Comparison of voltage distribution of the method and particle swarm optimization method at the stable moment of scenario 4

[0103] Fig.14 The method proposed in this invention is the Koopman linear optimization method which optimizes only the active power.

[0104] Comparison of current distribution between the method and particle swarm optimization method in scenario 1

[0105] Fig.15 This is a comparison diagram of current distribution of the method proposed in the present invention, the equal proportion allocation method, the Koopman linear optimization method that only optimizes active power, and the particle swarm optimization method in scenario 4.

[0106] Fig.16 This is a comparison of the current distribution of the method proposed in the present invention, the equal proportion allocation method, the Koopman linear optimization method that only optimizes active power, and the particle swarm optimization method at the stable moment of scenario 1.

[0107] Fig.17 This is a comparison of the current distribution of the method proposed in the present invention, the equal proportion allocation method, the Koopman linear optimization method that only optimizes active power, and the particle swarm optimization method at the stable moment of scenario 4. DETAILED DESCRIPTION

[0108] The present invention utilizes a large amount of distribution network operation measurement data, based on the Koopman dimensionality increase linearization method, and obtains a high-precision global power flow model through offline training. The online frequency modulation optimization layer uses the high-dimensional linear power flow model obtained by data drive to analyze the global sensitivity information online, and constructs a linear optimization model for primary frequency modulation of the active distribution network. The model is periodically rolled and solved, and the optimization results are continuously updated and sent to each local distributed power inverter. The local frequency response layer monitors the system frequency changes in real time, and immediately executes the corresponding instructions of each distributed power source when the frequency modulation startup requirements are met to ensure the rapidity of the frequency response. The present invention not only solves the problems of difficulty in optimization solution and heavy dependence on model parameters, but also improves the overall primary frequency modulation characteristics and power flow distribution of the distribution network. The present invention is described in detail below in conjunction with the accompanying drawings and embodiments.

[0109] 1 Global sensitivity modeling based on Koopman dimensionality-increasing linear mapping

[0110] The basic idea of ​​Koopman operator theory is to realize the linear evolution of nonlinear systems in finite state space in high-dimensional infinite space through the state space dimensionality mapping method. The linear power flow model based on Koopman data-driven mapping is essentially to collect a large amount of synchronous distribution network historical operation data, perform data-driven training in high-dimensional space through state space dimensionality transformation, and obtain a high-dimensional linear mapping relationship between input variables and output variables. Assume that the power flow relationship between the distribution network node voltage, line current, grid-connected point active power and node injection power can be expressed by the following nonlinear equations:

[0111] y=φ(x) (1)

[0112] Where: Input variable is x = [PQ] T , which is composed of active power and reactive power injected by the node, and the output variable y = [V IP pcc ] T It consists of node voltage, line current, and active power at the grid connection point. The purpose of training the linear mapping relationship between active power at the grid connection point and node power is to prepare for the subsequent establishment of a high-precision linear model of network loss under weak model conditions.

[0113] Using the Koopman operator to increase the dimension of the input variable is essentially a transformation and expansion of the coordinate system to obtain a high-dimensional global linear mapping model. lift represents the high-dimensional input variable after the state space is upgraded, and ψ(x) represents the dimensionality-enhancing operation function of the input variable x. Then there must be a linear matrix C that satisfies the linear mapping relationship between the input variable and the output variable:

[0114]

[0115] Theoretically, the above linear mapping relationship is strictly valid only when the Koopman operator is increased to infinite dimensions. However, in the actual training process, it is only necessary to increase the Koopman operator to a certain dimension to obtain a linear mapping model that meets the accuracy requirements. When the increased dimension is n, it can be regarded as a function of n scalar functions ψ i (x) constitutes, that is,

[0116] ψ i (x) = f lift (xc i ) (3)

[0117] Among them, c i is the i-th dimension basis vector randomly generated within the range of input variables; f lift It is a dimensionality-raising function. Generally speaking, there are many forms of dimensionality-raising functions. The present invention adopts a polyharmonic dimensionality-raising function with high computational efficiency, and its form is as follows:

[0118]

[0119] Among them, x i is the i-th element in x, c ij c i The jth element in , K is the dimension of the input variable x. After obtaining the historical operation data of the distribution network, the value of the linear power flow matrix C is estimated by the least squares method to obtain the dimension-increasing linear power flow equation. When there are S time-section historical data as the training sample set, the input variable sample set is The output variable sample set is Specifically, it can be expressed as

[0120]

[0121]

[0122] Based on the above samples, the linearized power flow matrix C can be obtained as follows:

[0123]

[0124] in, The Moore-Penrose inverse of the matrix represents the relationship between the output variable y and the input variable x. lift The linear power flow relationship between them is used to obtain a high-precision global power flow linear model. On this basis, the global sensitivity matrices X, Y, and Z corresponding to the node power of voltage, current, and grid-connected point active power can be further analyzed to obtain high-precision, thereby constructing a primary frequency modulation linear optimization model that does not depend on model parameters and realizing fast and accurate online control. They represent the voltage amplitude of node i, the current amplitude of the lth line (l∈E, indicating the branch number in the corresponding branch set), and the sensitivity of the active power of the grid connection point to the active power injected into node j. The global linear power flow matrix C obtained by Koopman data driving is The exact value of can be calculated according to formulas (8)-(10).

[0125]

[0126]

[0127]

[0128] Where N is the total number of nodes in the distribution network, l is the line number corresponding to branch ij, and C ij , C (N+l),j , C 2N,j Represent the corresponding output variables V in the linear power flow matrix C i ,I l , P PCC With input variable p j The element C i,(K+k) , C (N+l),(K+k) , C 2N,(K+k) Represent the corresponding output variables V in the linear power flow matrix C i ,I l , P PCC and the kth dimension-raising variable ψ k The partial differential terms on the right side of equations (8)-(10) can be calculated as follows:

[0129]

[0130] According to the above calculation process, the sensitivity matrix between the above three state variables and the node injected reactive power can be derived in the same way. Therefore, the sensitivity between any node voltage, line current, and grid-connected point active power and the node injected active and reactive power is The value of can be based on the linear power flow matrix C obtained by offline training, all the dimension-raising basis vectors c i , Node injected active power measurement j and reactive power measurement j It is found that the sensitivity information can be updated as the operating status changes. Therefore, the sensitivity calculation method is "global" and has strong nonlinear adaptability. This advantage will be reflected in the primary frequency modulation optimization control.

[0131] 2 Linear Optimization Model of Primary Frequency Regulation of Active Distribution Network Based on Koopman Global Sensitivity

[0132] By utilizing the global sensitivity obtained by data driving, a distribution network primary frequency regulation optimization model that is independent of network parameters is constructed, avoiding the complex mechanism analysis and modeling process, and converting the nonlinear optimization problem into a linear optimization problem, greatly reducing the computational burden.

[0133] 1) Objective function

[0134] Based on the global sensitivity of the active power of the grid-connected point to the node power, the absolute value of the difference between the total net load active power and the actual active power of the grid-connected point is used to indirectly represent the size of the network loss, thereby establishing a high-precision linearization model of network loss that is independent of network parameters. The specific objective function is as follows:

[0135]

[0136] The first term of the objective function represents the network loss, and the second term represents the voltage distribution. Among them, Γ represents the set of nodes connected to photovoltaics, Π represents the set of nodes connected to energy storage, Υ represents the set of nodes connected to the load, represents the initial active power of the photovoltaic at node i, represents the initial active power of energy storage at node i, Represents the load active power at node i. They respectively represent the active regulation of photovoltaic and energy storage at node i at time t. It represents the active power of the grid connection point at time t.

[0137] During a frequency regulation period, the active regulation of the distribution network is determined by the total frequency regulation droop coefficient and the system frequency deviation issued by the main grid operator, which can be expressed as follows:

[0138]

[0139] Since the total droop coefficient at each optimization moment Deviation from system frequency Δf t are all constants, so the corresponding active power regulation of the distribution network is also a fixed constant. The objective function (12) can be further simplified as follows:

[0140]

[0141] 2) Power flow constraints

[0142] According to the global power flow sensitivity X, Y, Z obtained by data driving, the power flow and safety constraints of the distribution network can be expressed by the following linear relationship:

[0143]

[0144] In the formula, They represent the vectors composed of the voltage amplitudes of each node and the line current amplitudes in the initial state respectively; Indicates the active power of the grid connection point in the initial state; X P ∈R N×N , X Q ∈R N×N are the voltage-node active sensitivity mapping matrix and the voltage-node reactive sensitivity mapping matrix, Y P ∈R (N-1)×N , Y Q ∈R (N-1)×N are the current-node active sensitivity mapping matrix and the current-node reactive sensitivity mapping matrix respectively; Z P ∈R 1×N , Z Q ∈R 1×N They are the grid-connected point active power-node active power sensitivity mapping matrix and the grid-connected point-node reactive power sensitivity mapping matrix; ΔP t , ΔQ t They represent the vectors composed of the active and reactive regulation quantities of each active node at time t; R and S represent the node position matrices of photovoltaic and energy storage in the distribution network respectively; They represent the vectors composed of the active changes of photovoltaic power and energy storage at time t respectively; They represent the vectors composed of the reactive power changes of photovoltaic and energy storage at time t respectively.

[0145] In summary, formulas (1)-(15) construct a complete linear primary frequency regulation optimization control model for active distribution networks. In actual operation control, the linearized power flow matrix C obtained through offline model training can be used as a fixed parameter and only needs to be checked periodically. In frequency regulation optimization control, the power flow sensitivity information is calculated online based on the linearized power flow matrix C and the current operating status. It is substituted into and solved in the linear optimization control model to obtain the droop coefficient and reactive power regulation instruction of each distributed power source. When the operating state of the distribution network changes, the sensitivity information and the optimization control model are updated online periodically to achieve online frequency regulation optimization control. In the local frequency response, each distributed power source only needs to respond to the system frequency change in real time according to the received frequency regulation instruction to achieve a specific overall rapid frequency response of the distribution network. In the entire operation control, since the sensitivity calculation itself does not need to be repeatedly data-driven training, and the optimization model is an easy-to-solve linear model, it can meet the requirements of rapid dynamic control of the primary frequency regulation of the distribution network. The overall process of active distribution network primary frequency regulation control based on Koopman global data-driven sensitivity is as follows. Figure 1 shown.

[0146] 3 Case Analysis

[0147] In order to verify the effectiveness and accuracy of the present invention, MATLAB software was used to build a 10kV virtual medium-voltage distribution network with high penetration distributed photovoltaic and energy storage. The topology is as follows: Figure 2 The total installed capacity of distributed photovoltaic and energy storage is approximately 4.68MW and 3.4MW respectively. The specific control parameters are shown in Table 1. The rated voltage and rated current of the distribution network are set as the reference voltage and reference current, the first node voltage is 1p.u, the upper and lower limits of the voltage operation are 1.07pu and 0.93pu respectively, and the upper limit of the line current operation is 0.135pu.

[0148] Table 1

[0149]

[0150] Since the frequency modulation of the distribution network involves various working conditions such as active reverse transmission and absorption, frequency rise and drop, etc., in order to verify that the present invention has good optimization effects under different working conditions, four different scenarios are set for simulation verification: 1) The frequency drops when the distribution network reverses active power. 2) The frequency increases when the distribution network reverses active power. 3) The frequency drops when the distribution network absorbs active power. 4) The frequency increases when the distribution network absorbs active power. Among them, the total net load active power corresponding to the reverse transmission and absorption of active power of the distribution network is -2.5299MW and 2.1649MW respectively.

[0151] In order to verify the superiority of the method of the present invention, the control effects of the method are compared with those of the proportional allocation (PA), the Koopman linear optimization (KLO-AP) that optimizes only active power, and the particle swarm optimization (PSO) in four different scenarios. The comparison factors of the four control methods include optimization results, frequency modulation external characteristics, and voltage and current limit violations. Among them, the particle swarm algorithm (PSO) is a standard form. In order to enable the algorithm to achieve better global and local search capabilities, the inertia weight w adopts a linear decreasing strategy, and after multiple experiments and experience summarization, the appropriate parameter settings are determined, specifically: the maximum inertia weight w max =0.9, minimum inertia weight w min =0.4, the particle swarm size N=500, and the learning factors s1 and s2 are 0.5 and 0.9 respectively.

[0152] Example 1

[0153] Figure 3 , Figure 4 and Figure 5The optimization results of four frequency modulation control methods in four different scenarios are given. It can be seen that in terms of network loss, total voltage deviation and objective function (the sum of network loss cost and voltage deviation cost), the Koopman linear optimization (KLO) proposed in the present invention has the best comprehensive control effect in the four scenarios, followed by PSO, KLO-AP, and PA control effect is the worst. After calculation, taking PA as a reference, in a frequency modulation process, KLO reduces network loss by an average of 56.10%, total voltage deviation by an average of 69.02%, and objective function value by an average of 68.76%, which is superior to the other three methods in terms of loss reduction and voltage reduction.

[0154] Example 2

[0155] Table 2 shows the comparison of KLO and KLO-AP in terms of computing time. It can be seen that in the four scenarios, the KLO proposed in the present invention is far superior to POS in terms of computing time. The main reason is that POS needs to update the particle position and velocity for multiple iterations when searching the solution space. The amount of calculation for each iteration is large, and the original frequency modulation optimization model is high-dimensional nonlinear and difficult to solve. It requires a large number of iterations, resulting in a long optimization calculation time. KLO uses Koopman dimensional linearization to calculate the global tidal sensitivity, converting the original nonlinear optimization model that is difficult to solve into a linear optimization model that is easy to solve, greatly speeding up the optimization calculation time.

[0156] Table 2

[0157]

[0158] Example 3

[0159] Figure 6 and Figure 7 The power-frequency characteristic curve results of KLO and other three control methods in scenario 1 and scenario 3 are given. For the convenience of comparison, the preset ideal power-frequency characteristic curve is also Figure 6 It is given in Figure 2 that the degree of closeness between the power-frequency characteristic curves under different control methods and the ideal power-frequency characteristic curve depends on the respective loss reduction effects. It can be seen that the overall power-frequency characteristic curve of the distribution network under KLO control is closest to the ideal power-frequency characteristic curve, followed by PSO, then KLO-AP, and the farthest is PA.

[0160] Figure 8 and Fig. 9 The frequency change curves of the four control methods in scenario 1 and scenario 3 are given respectively. It can be seen that since the frequency modulation characteristics are closest to the ideal curve, KLO has the best frequency stability effect, while PA has the worst frequency stability effect. Compared with PA, KLO improves the frequency stability value by 0.0161Hz and 0.0057Hz in scenario 1 and scenario 3 respectively.

[0161] Example 4

[0162] Fig.10 and Fig.11 The voltage distribution and dynamic changes of the four control methods in the frequency modulation process under scenario 1 and scenario 4 are given. It can be seen that under KLO control, the voltage distribution in the frequency modulation process is more balanced, which greatly reduces the voltage deviation cost and solves the voltage over-limit problem in the frequency modulation process. In contrast, under PA control, the node voltage in scenario 1 is greatly increased by the frequency modulation service that suppresses the frequency drop, and the node voltage in scenario 4 is greatly reduced by the frequency modulation service that suppresses the frequency rise. The voltage fluctuations are large, threatening the safe operation of the distribution network. In addition, in the control based on KLO-AP, an attempt is made to reduce the voltage deviation cost in the frequency modulation process by optimizing the droop coefficient of each DER in the distribution network, but the voltage reduction effect is not significant, and the voltage still has large fluctuations under the influence of the frequency modulation power. At the same time, under POS control, although the voltage fluctuations affected by the frequency modulation power in different scenarios are effectively alleviated and the voltage deviation is reduced, the optimal effect is not achieved, and there is a gap with the voltage control effect of KLO.

[0163] In order to more intuitively reflect the voltage limit situation, Fig.12 and Fig.13 The voltage distribution at a certain moment after the system frequency stabilizes is given. It can be seen that in scenarios 1 and 4, PA and KLO-AP cause the voltage to exceed the upper and lower limits respectively due to their unreasonable and insufficient frequency modulation power allocation. Both PSO and KLO can effectively solve the voltage limit problem during the frequency modulation process, and the voltage distribution of KLO is more balanced.

[0164] Example 5

[0165] Fig.14 and Fig.15 The line current distribution and dynamic changes of the four control methods in the frequency modulation process under scenario 1 and scenario 4 are given. It can be seen that under KLO control, the current of the distribution network in the frequency modulation process is effectively reduced and the distribution is more balanced. In contrast, under PA and KLO-AP control, the line current has large fluctuations due to the influence of the frequency modulation power, and the trunk line current is always large, affecting the stable, safe and economical operation of the distribution network. In addition, POS can reduce the line current in the frequency modulation process to a certain extent, but it does not achieve the optimal effect, and there is a gap with the current control effect of KLO.

[0166] In order to more intuitively reflect the current limit situation, Fig.16 and Fig.17The current distribution at a certain moment after the system frequency stabilizes is given. It can be seen that under the control of PA and KLO-AP, current limit violations occur in both scenario 1 and scenario 4. In contrast, both PSO and KLO can effectively solve the current limit violation problem during the frequency modulation process.

Claims

1. A method for optimizing and coordinating primary frequency regulation of an active distribution network under weak model conditions, comprising the following steps: (1) Construct a power flow model with Koopman state space dimension increase; (2) Construct an active distribution network primary frequency regulation optimization control model. Its optimization goal is to optimize the droop coefficient and reactive power regulation of DER in multiple time periods, aiming to minimize the algebraic sum of network loss and voltage deviation under the premise of satisfying the overall frequency regulation characteristics and power flow constraints. Its constraints are: the time-series coupled algebraic equation of the system frequency response is obtained through differential discretization. When the frequency deviation exceeds the dead zone, each distributed power source in the distribution network provides active power support through its own droop characteristics. The voltage amplitude of each node and the current amplitude of each line are calculated to determine whether the power flow inside the distribution network is within a safe and reasonable range to avoid voltage and current exceeding the limit. The power of each photovoltaic unit needs to meet the capacity limit, and each energy storage unit needs to meet the power and power capacity constraints. (3) Solve the established primary frequency regulation optimization control model to realize the primary frequency regulation optimization coordinated control of the active distribution network taking into account the power flow safety constraints under weak model conditions.

2. The method for optimizing and coordinating primary frequency regulation of active distribution network under weak model conditions according to claim 1 is characterized in that: The method of step (1) is as follows: The basic form of the power flow equation is as follows: Among them, the input variable is x = [PQ] T , which is composed of active power and reactive power injected by the node, and the output variable y = [VI P pcc ] T It is composed of node voltage, line current, and active power at the grid connection point. C represents the linearized power flow mapping matrix, x lift represents the high-dimensional input variable after the state space is upgraded, and ψ(x) represents the dimensionality-elevating operation function of the input variable x; When the dimension is increased to n, ψ(x) is composed of n scalar functions ψ i (x) constitutes, namely: ψ i (x)=f lift (x-c i ) (2) Among them, f lift represents the expanded form of the dimension-raising function, c i represents the expanded i-th dimension basis vector, whose value is selected as a random value within the range of the distribution network input variable, c ij Represents c i The jth element in , K is the dimension of the input variable x; When there are S time-section historical data as training sample sets, the input variable sample set is The output variable sample set is Based on the historical operation data of the distribution network, data-driven training is carried out, and the linearized power flow mapping matrix C can be obtained through least squares estimation. The calculation formula is as follows: in, represents the Moore-Penrose inverse of a matrix; make They represent the voltage amplitude of node i, the current amplitude of the lth line (l∈E, indicating the branch number in the corresponding branch set), and the sensitivity of the active power of the grid connection point to the active power injected into node j; the linearized power flow mapping matrix C driven by Koopman data, the sensitivity The calculation formulas are as follows: Among them, V i ,I l , P PCC They are respectively represented as node voltage, line current and active power at the grid connection point, N is the total number of distribution network nodes, C ij , C (N+l),j , C 2N,j Represent the corresponding output variables V in the linearized power flow mapping matrix C i ,I l , P PCC With input variable p j The element C i,(K+k) , C (N+l),(K+k) , C 2N,(K+k) Represent the corresponding output variables V in the linearized power flow mapping matrix C i ,I l , P PCC and the kth dimension-raising variable ψ k (x) is an element; the partial differential term on the right side of the equations (5)-(7) can be calculated as follows: According to the above calculation process, the node active injection power in the denominator of equations (5)-(8) is replaced by the node reactive power, and the above three state variables V are derived in the same way. i ,I l , P PCC Sensitivity matrix with reactive power injected into the node.

3. The method for optimizing and coordinating primary frequency regulation of active distribution network under weak model conditions according to claim 2 is characterized in that: In step (2), the optimization objective function established is: The optimization objective function can be expressed as: ΔQ=(ΔQ 1 ,…,ΔQ t ,…,ΔQ T ) (11) Among them, the first term of the objective function represents the network loss, and the second term represents the voltage distribution; C1, C2 are weight coefficients, t is the time point in the optimization period, T is the number of time points in the optimization period, N is the total number of distribution network nodes, B represents the total number of optimized particles of DER reactive regulation in the optimization period, the symbol < > represents rounding up, n and m represent the number of photovoltaic and energy storage included in the distribution network, respectively, Γ represents the set of nodes connected to photovoltaic, Π represents the set of nodes connected to energy storage, Υ represents the set of nodes connected to the load, represents the initial active power of the photovoltaic at node i, represents the initial active power of energy storage at node i, represents the load active power at node i, They represent the active regulation of photovoltaic power and energy storage at node i at time t respectively; Represents the active power of the grid connection point at time t, V i t represents the node voltage, K f represents the (m+n)×1-order column vector consisting of the droop coefficients of the independent variable DER, ΔQ t and The (m+n)×1-order column vectors correspond to the reactive power regulation of DER at the t-th time and the k-th optimization granularity. ΔQ represents the reactive power dispatch sequence of each DER in the optimization period, which is composed of T ΔQ t A (m+n)×T matrix.

4. The method for optimizing and coordinating primary frequency regulation of active distribution network under weak model conditions according to claim 3 is characterized in that: The constraints in step (2) are specifically: The time-series coupled algebraic equations of the system frequency response obtained by differential discretization are as follows: Among them, t1 represents the frequency discrete point, Δt1 represents the frequency discrete interval, f represents the system real-time frequency, H represents the power system inertia coefficient, P0 represents the initial active output of the distribution network, P G Represents the active output of the conventional electric field, P L Indicates the active load of the power grid, K G and K f They represent the droop coefficients of conventional power stations and distribution networks respectively; When the frequency deviation exceeds the dead zone, each distributed power source in the distribution network provides active power support through its own droop characteristics, as shown in the following formula: in, and Represent the active power of photovoltaic and energy storage respectively, and Represent the active initial power of photovoltaic and energy storage respectively, and are the droop coefficients of PV and energy storage, respectively; Calculate the voltage amplitude of each node and the current amplitude of each line to determine whether the internal power flow of the distribution network is within a safe and reasonable range to avoid voltage and current exceeding the limit, as shown in the following formula: In the formula, They represent the vectors composed of the voltage amplitudes of each node and the line current amplitudes in the initial state respectively; Indicates the active power of the grid connection point in the initial state; X P ∈R N×N , X Q ∈R N×N are the voltage-node active sensitivity mapping matrix and the voltage-node reactive sensitivity mapping matrix, Y P ∈R (N-1)×N , Y Q ∈R (N-1)×N are the current-node active sensitivity mapping matrix and the current-node reactive sensitivity mapping matrix respectively; Z P ∈R 1×N , Z Q ∈R 1×N They are the grid-connected point active power-node active power sensitivity mapping matrix and the grid-connected point-node reactive power sensitivity mapping matrix; ΔP t , ΔQ t They represent the vectors composed of the active and reactive regulation quantities of each active node at time t; R and S represent the node position matrices of photovoltaic and energy storage in the distribution network respectively; They represent the vectors composed of the active changes of photovoltaic power and energy storage at time t respectively; They represent the vectors composed of the reactive power changes of photovoltaic and energy storage at time t respectively; The power of each photovoltaic unit needs to meet the capacity limit, which is expressed as follows: in, represents the maximum active output of the i-th photovoltaic unit; S i,pv is the rated capacity of the corresponding ith PV unit; Each energy storage unit must meet the power and power capacity constraints, as shown in the following formula: in, is the energy storage capacity connected to the i-th node at time t; and are the corresponding energy storage charging power and discharging power respectively; η charge and η discharge are the charge and discharge efficiencies respectively; Δt is the optimization time interval; and They are the upper limits of the charging and discharging power of energy storage respectively; Indicates the charge and discharge mark of energy storage, and introduces Constraint means that at any time, the energy storage can only be in one of the three states: charging, discharging, or neither charging nor discharging. It is the energy storage capacity limit.

Citation Information

Cited By

  • Power system steady-dynamic resource scheduling method considering frequency modulation capability dependency

    CN121906452A