KR102406851B1 - Coordinated optimization method for maximizing the power of wind farm using scalable wake digraph and apparatus performing the same - Google Patents

Abstract

According to the present invention, a method for optimizing power regulation of a wind farm (WF) includes: collecting parameters for the plurality of wind turbines (WT), which are associated with output power of the wind farm; generating a wake weighting coefficient matrix numerically representing a wake effect acting between the plurality of WTs, and a wake direction graph representing a wake effect acting between the WTs as a directional line using the WT as a node; performing a pruning process on the wake weighting coefficient matrix and the wake direction graph using a threshold defined based on a wake weighting coefficient; classifying the plurality of WTs into a plurality of subsets by clustering the plurality of WTs based on a wake weighting coefficient matrix and a wake direction graph on which the pruning process is completed; determining a sparse communication architecture applied to the plurality of WTs based on the subset classification result; and generating a power adjustment optimization solution capable of adjusting a yaw angle and an axis coefficient of each WT by applying a predefined power adjustment optimization algorithm corresponding to the determined sparse communication architecture.

Description

A method for optimizing power adjustment of a wind farm using a scalable wake direction graph and an apparatus for performing the same

The present invention relates to an optimization technique for realizing the maximum power efficiency of a wind farm composed of a plurality of wind turbines.

A wind farm composed of a plurality of wind turbines has a limit in securing maximum power due to the effect of a wake effect. More specifically, a wind turbine located downstream can produce less power than when operated in a free stream due to the wind speed deficit by the wind turbine upstream. This wake effect has a great influence on the output power of the wind farm.

Various techniques for optimizing power in consideration of the wake effect have been proposed. However, in the case of the prior art, the power increase rate is limited to about 5% or less, and it is very inefficient in terms of information exchange and data processing.

Currently, as a strategy for optimizing the power adjustment of a wind farm in consideration of the wake effect, it is necessary to develop a technology capable of further improving power efficiency and further improving information exchange and data processing efficiency.

Korean Patent Publication No. 10-2013-0039156

The present invention has been derived to solve the above-described problems, and a method for optimizing power adjustment of a wind farm using an expandable wake direction graph, which can further improve power efficiency, information exchange and data processing efficiency, and a method for performing the same We want to provide a device.

According to an aspect of the present invention, a method for optimizing power adjustment of a wind farm (WF) includes: collecting parameters for the plurality of wind turbines (WT), which are associated with output power of the wind farm; generating a wake weighting coefficient matrix numerically representing a wake effect acting between the plurality of WTs, and a wake direction graph representing a wake effect acting between the WTs as a directional line using the WT as a node; performing a pruning process on the wake weighting coefficient matrix and the wake direction graph using a threshold defined based on a wake weighting coefficient; classifying the plurality of WTs into a plurality of subsets by clustering the plurality of WTs based on a wake weighting coefficient matrix and a wake direction graph on which the pruning process is completed; determining a sparse communication architecture applied to the plurality of WTs based on the subset classification result; and generating a power adjustment optimization solution capable of adjusting a yaw angle and an axis coefficient of each WT by applying a predefined power adjustment optimization algorithm corresponding to the determined sparse communication architecture.

In an embodiment, performing the pruning process may include calculating an average value for all wake weighting coefficients, and defining the calculated average value as the threshold value. Here, the performing of the pruning process may further include adjusting the defined threshold based on a hyper parameter that controls a sparse degree of the wake direction graph.

In an embodiment, the sparse communication architecture may include a decentralized communication architecture and a distributed communication architecture.

In one embodiment, determining the sparse communication architecture comprises: determining whether a shared WT exists between the classified subsets; and determining as a non-centralized communication architecture if the shared WT does not exist, and determining as a distributed communication architecture if the shared WT exists.

In an embodiment, the generating of the power adjustment optimization solution includes applying a predefined decentralized optimization algorithm when the applied sparse communication architecture is determined as a non-centralized communication architecture to obtain the power adjustment optimization solution. and if the applied sparse communication architecture is determined to be a distributed communication architecture, a predefined distributed optimization algorithm may be applied to generate the power adjustment optimization solution.

In accordance with another aspect of the present invention, there is provided an apparatus for optimizing power adjustment of a wind farm, comprising: a communication architecture determining unit configured to determine a communication architecture applied to the plurality of WTs; and a power adjustment optimization unit for generating a power adjustment optimization solution capable of adjusting the yaw angle and axis coefficient of each WT by applying a predefined power adjustment optimization algorithm corresponding to the determined communication architecture. Here, the communication architecture determining unit collects parameters for the plurality of WTs, which are related to the output power of the wind farm, and a wake weighting value that numerically represents a wake influence between the plurality of WTs. A coefficient matrix, the WT as a node, and a wake-direction graph representing a wake effect acting between the WTs as a direction line are generated, and using a threshold defined based on a wake weighting coefficient, the wake weighting coefficient matrix and A pruning process is performed on the wake direction graph, and the plurality of WTs are clustered based on the wake weighting coefficient matrix and the wake direction graph on which the pruning process is completed and classified into a plurality of subsets, and the subset A sparse communication architecture applied to the plurality of WTs is determined based on the classification result.

In an embodiment, the communication architecture determiner may calculate an average value for all wake weighting coefficients, and define the calculated average value as the threshold value. Here, the communication architecture determining unit may adjust the defined threshold based on a hyper parameter that controls a sparse degree of the wake direction graph.

In an embodiment, the sparse communication architecture may include a decentralized communication architecture and a distributed communication architecture.

In an embodiment, the communication architecture determining unit determines whether a shared WT exists between the classified subsets, and if the shared WT does not exist, determines a non-centralized communication architecture, and if the shared WT exists It can be decided by a distributed communication architecture.

In one embodiment, when the applied sparse communication architecture is determined to be a non-centralized communication architecture, the power adjustment optimization unit generates the power adjustment optimization solution by applying a predefined decentralized optimization algorithm, When the applied sparse communication architecture is determined as the distributed communication architecture, a predefined distributed optimization algorithm may be applied to generate the power adjustment optimization solution.

The present invention determines a communication architecture applied to a plurality of WTs based on a scalable wake direction graph, and applies a predefined power adjustment optimization algorithm corresponding to the determined communication architecture, yaw angle and axis coefficient of each WT By creating a power adjustment optimization solution that can adjust the power efficiency, information exchange and data processing efficiency can be further improved.

1 to 10 are reference diagrams for explaining a method of optimizing power adjustment of a wind power farm using an extendable wake direction graph according to the present invention.
12 is a block diagram for explaining an apparatus for optimizing power adjustment of a wind power farm using a scalable wake direction graph according to the present invention.

Since the present invention can apply various transformations and can have various embodiments, specific embodiments are illustrated in the drawings and described in detail in the detailed description. However, this is not intended to limit the present invention to specific embodiments, and should be understood to include all modifications, equivalents, and substitutes included in the spirit and scope of the present invention. In describing the present invention, if it is determined that a detailed description of a related known technology may obscure the gist of the present invention, the detailed description thereof will be omitted.

Terms such as first, second, etc. may be used to describe various elements, but the elements should not be limited by the terms. The above terms are used only for the purpose of distinguishing one component from another.

The terms used in the present application are only used to describe specific embodiments, and are not intended to limit the present invention. The singular expression includes the plural expression unless the context clearly dictates otherwise. In the present application, terms such as “comprise” or “have” are intended to designate that a feature, number, step, operation, component, part, or combination thereof described in the specification exists, but one or more other features It should be understood that this does not preclude the existence or addition of numbers, steps, operations, components, parts, or combinations thereof. Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings.

1. Introduction

Coordinated control of wind power plants is critical to improving power production in off-shore wind farms (OWFs), reducing turbine fatigue loads in OWFs, and tracking power reference signals in energy grids. In general, conventional greedy control operates individually to maximize the performance of all turbines, and does not take into account the wake effect of adjacent turbines. However, modern OWFs consist of many turbines, and the wake effect cannot be neglected. Due to the influence of wake effect on power production, there is a need to derive an efficient communication architecture and a coordinated control strategy.

Communication architectures for large OWF-level coordinated controllers are centralized, distributed, and decentralized. In general, wind turbine (WT) control systems employ a centralized communication architecture that relies on a centralized controller to optimize decisions for all turbines. However, in large-scale OWF, the number of control parameters for WF increases exponentially as the number of turbines increases, which is inefficient in terms of cost and calculation.

To alleviate these shortcomings, a sparse communication architecture should be built within a large OWF cluster. Here, clustering adjacent turbines into subsets taking into account the wake effect would be key. The sparse communication architecture requires limited neighbor communication to realize optimal control of the WF. In related studies, in order to achieve a sparse communication architecture, each WF level control device allows communication between WTs in a direct neighbor relationship, or extends to an arbitrary set of neighbors, or a single downstream or upstream neighbor. A method of limiting it has been proposed. However, these methods of determining the communication neighbor only consider the physical distance and there are no specific rules. In other prior studies, it has been proposed to determine the communication neighbor of the WT by calculating the horizontal and longitudinal distances of the WT, or to group the WT according to the average velocity deficit factor of the downstream by the upstream WT. However, this method only solves a specific wind direction problem and cannot solve a complex wind direction problem. Meanwhile, in another prior study, a distributed communication architecture approach in which communication neighbors are determined by grid division has been proposed. However, this method has limitations in a way that considers the interrelationship for electrical connection rather than wake effect.

The present invention has been devised to overcome the limitations of the prior studies, and is coordinated to build a sparse communication architecture through a scalable wake directed graph or digraph and maximize the power production of OWF. optimization) is proposed.

More specifically, the present invention proposes a scalable wake-directed graph that can define centralized, distributed, and decentralized communication architectures.

In addition, the present invention proposes various coordination optimization algorithms consistent with the communication network topology to solve the non-convex optimization problem.

In addition, the present invention proposes a practical methodology for setting up a sparse communication architecture based on a scalable wake direction graph and weight coefficients. This methodology can reduce communication neighbors through an adaptive k-threshold pruning algorithm. Meanwhile, communication neighbors may be clustered through connectivity analysis of the sparse wake direction graph. Here, each WT may be assigned multiple leaders or a single leader.

Hereinafter, the adjustment optimization technique of the wind power farm according to the present invention will be described in more detail with reference to drawings and equations.

2. Wake model considering yaw angle

The flow correlation between the WTs depends on the wind direction and wind speed of the incoming wind, which changes the wake overlap between the WTs and affects the wake distribution characteristics of the WFs. Accordingly, the accurate description of the wake distribution of the WT is the basis for realizing the clustering of the WT and determining the communication neighbor.

In the present invention, the aerodynamic wake interaction is described using the two-dimensional WF wake model, that is, the FLORIS model, and the performance of the WF control algorithm is evaluated. First, in order to explain the spatial characteristics of the wake behind the WT, a Cartesian coordinate system (x,y) is used in which the X-axis points to the wind inflow direction and the Y-axis points to a direction orthogonal to the X-axis as shown in FIG. 1A. The headwind-crosswind coordinates can be expressed as the following equation.

[Formula (1)]

The FLORIS model mainly consists of wake deficit, wake defection, and wake expansion models, and there are also models that calculate the wake overlap area, the inflow wind speed of the downstream WT, WF power, thrust, etc. may include Here, the process of calculating the inflow wind speed of the downstream WT related to the present invention will be mainly described.

In the FLORIS model, three wake zones can be defined. Specifically, wake zones are a near wake (q = 1), a far wake (q = 2), and a mixed zone (q = 3). There is (see Fig. 1b). By combining the effects of wake zones of the upstream WTi, the effective speed of the downstream WTj can be expressed as the following equation.

[Formula (2)]

는 WF의 방해받지 않는 유입풍속,

는 업스트림 WT의 축 유도계수(axial induction factor),

는 j WT의 웨이크존 q 속도결손계수(velocity deficit factor),

는 i WT q 웨이크 존과 j WT 로터의 오버래핑 영역(area), Aj는 WT 로터의 다운스트림 WT 로터 면적, (Xi, Yi)는 WT i 와 j의 좌표임)(where Uj is the effective inlet wind speed of WTj,

is the undisturbed inflow wind speed of the WF,

is the axial induction factor of the upstream WT,

is the wake zone of j WT, q velocity deficit factor,

where i WT q wake zone and j overlapping area of the WT rotor, Aj is the area of the WT rotor downstream of the WT rotor, (Xi, Yi) is the coordinates of WT i and j)

[Equation (3)]

는 웨이크 확장 계수(wake expansion coe-cients),

는 업스트림 WTi의 요 각도, Xj - Xi는 다운스트림 WTj와 업스트림 WTi 사이의 상대 거리)(where Di is the WTi rotor diameter, ke and

is the wake expansion coe-cients,

is the yaw angle of the upstream WTi, Xj - Xi is the relative distance between the downstream WTj and the upstream WTi)

[Formula (4)]

,

,

는 각각 웨이크 모델 상수이다. 도 1a에서, 업스트림 WTi의 근거리 웨이크존

과 다운스트림 WT j의 로터 사이에 중첩이 없으므로,

이다. 유효 풍속 계산될 때, 다운스트림 WT j의 전력 함수(power function)는 은 다음 식과 같이 표현될 수 있다. From here,

,

,

are the wake model constants, respectively. 1A, the near wake zone of the upstream WTi

and there is no overlap between the rotor of the downstream WT j,

to be. When the effective wind speed is calculated, the power function of the downstream WT j can be expressed as the following equation.

[Formula (5)]

는 발전기 효율(generator e-ciency),

는 공기 밀도, Aj는 로터 스윕 영역(swept area),

는 요각이며, Uj는 유효 풍속을 결정하고,

는 축 유도 계수(axial induction factor)이고,

는 요 비정렬의 보정 계수이며, Pp는 요 비정렬로 인한 전력 손실에 일치하는 튜닝 가능한 파라미터를 의미하며, 본 발명에서는 Pp = 1.88로 정의된다. here

is the generator e-ciency,

is the air density, Aj is the rotor sweep area,

is the yaw angle, Uj determines the effective wind speed,

is the axial induction factor,

is a correction coefficient of yaw misalignment, Pp is a tunable parameter corresponding to power loss due to yaw misalignment, and is defined as Pp = 1.88 in the present invention.

3. Building a sparse communication architecture using a scalable directed graph

The first step in building a centralized, distributed, and decentralized coordination controller is to derive a communication architecture corresponding to each. The present invention proposes a practical approach to building a sparse communication architecture through a scalable wake directed graph. This will be briefly described with reference to FIG. 2 .

1) Wake field

Referring to FIG. 2 , the wake propagation characteristic of the wake field may be described based on the FLORIS wake model. When different wind directions and wind speeds flow through the WF, the upstream WT absorbs kinetic energy in the air. As a result, the inlet wind speed of the downstream turbine is reduced, and wake interference is increased.

2) Wake digraph

The wake graph is directional and has a fixed communication architecture. Wake propagation information is exchanged unidirectionally from upstream to downstream, and it can be viewed as a graph having directionality. The arrows in FIG. 2 indicate the interaction with the direction due to the wake between the turbines.

3) Pruning and clustering

Based on the wake directed graph, a sparse turbine communication network architecture can be built using pruning and clustering algorithms. This is to reduce the number of turbines in the neighboring telecommunication neighbors.

4) Communication architecture

Referring to FIG. 2 , a control communication architecture for a large OWF cluster including a plurality of turbines may be classified into a centralized, decentralized, and distributed control communication architecture. . In the case of the centralized control method, the cooperative action action is instructed by the central control unit (see equation (8)). In the case of a distributed control scheme, cooperative action actions may be processed locally or remotely (see equation (10)) by a central control unit. In the case of a decentralized type, it can be controlled by individual independent controllers rather than a single controller. Here, an appropriate optimization algorithm corresponding to each communication architecture should be selected.

3.1. wake direction graph

를 정의한다. 여기서, 꼭지점

는 WT(개별 에이전트)이고, 엣지

는 WT간(연결된 토폴로지) 웨이크 분포를 나타내며, 웨이트(weights)

는 웨이크 강도 계수(통신 규약)를 추상화한다. 또한, 노드 i에 대한 인접 노드 집합은

로 정의된다. Ni 하위 집합(subset)과 다른 하위 집합 Nj 사이의 공유 노드는

로 정의되고, 여기에서, Ti는 WF의 터빈 번호를 의미한다. 음이 아닌 행렬

는 인접 가중 행렬(adjacency weighting matrix)로서 웨이크 강도 가중 행렬(wake intensity weighting matrix)이며, aij는 WTj에 작용하는 WTi의 웨이크 가중치 값이다.Assume that there are n WTs in the WF, and the wake direction graph

to define Here, the vertex

is the WT (individual agent), edge

represents the wake distribution between WTs (connected topology), and the weights

abstracts the wake strength factor (communication protocol). Also, the set of adjacent nodes for node i is

is defined as A shared node between a subset of Ni and another subset Nj is

, where Ti means the turbine number of WF. nonnegative matrix

is a wake intensity weighting matrix as an adjacency weighting matrix, and aij is a wake weight value of WTi acting on WTj.

[Equation (6)]

The wake weighting factor is aij if there is a wake effect of WTi on WTj, and each item is 0 if there is no wake effect between each other. Therefore, the wake weighting coefficient of aij can be calculated according to the following equation.

[Equation (7)]

, 속도 결손(velocity deficit)

및 물리적 거리 x가 고려될 수 있다. 여기에서,

은 WT j의 로터 위험(rotor risk)에 대한 WTi 웨이크의 오버랩 비율이며,

는 업스트림 WT i의 웨이크에 의한 다운스트림 WT j의 결손 인자(deficit factor)이고,

는 WTi 및 WTj 간의 물리적 거리이며, D는 WT 로터 직경이다. 한편, 모든 WT의 직경이 동일한 것으로 가정한다.In Equation (7) above, there are three factors to define the wake weighting factor aij, the wake overlap.

, velocity deficit

and physical distance x can be considered. From here,

is the overlap ratio of WTi wake to rotor risk of WT j,

is the deficit factor of the downstream WT j by the wake of the upstream WT i,

is the physical distance between WTi and WTj, and D is the WT rotor diameter. On the other hand, it is assumed that all WTs have the same diameter.

3.2. Building a sparse communication architecture through clustering and pruning

를 기반으로, 적응형 k 임계 가중치 그래프(adaptive k-threshold weight graph) 가지치기 및 연결 기반(connectivity-based) 그래프 클러스터링 알고리즘을 이용하여 희소 통신 아키텍처를 희소하게(sparse) 할 수 있다. 적응형 k 임계

가지치기는 일정 값 이하의 웨이크 가중 계수를 제거함으로써 웨이크 방향 그래프의 사이즈를 축소할 수 있다. 여기에서, 계수 k는 웨이크 방향 그래프의 희소 정도(sparse degree)를 제어하는 하이퍼 파라미터에 해당하며, 임계값

는 전체 웨이크 가중치 계수의 평균 값과 같을 수 있다(예 :

). 방향 그래프의 희소화 작업(sparseness)이 완료된 후 WT의 클러스터링 작업이 시작된다. 여기에서, 연결성(connectivity), 퍼지 클러스터링(fuzzy clustering) 및 k-평균(k-means) 방법 등 실용적인 방향 그래프 클러스터링 방법론이 적용될 수 있으며, 이하에서는 연결성(connectivity) 방법론을 예시로 하여 설명한다. original wake direction graph

Based on , it is possible to sparse a sparse communication architecture using an adaptive k-threshold weight graph pruning and a connectivity-based graph clustering algorithm. Adaptive k-threshold

The pruning may reduce the size of the wake direction graph by removing a wake weighting coefficient less than or equal to a predetermined value. Here, the coefficient k corresponds to a hyperparameter controlling the sparse degree of the wake-directed graph, and a threshold value

may be equal to the average value of all wake weighting coefficients, e.g.:

). After the sparseness of the directed graph is completed, the clustering operation of the WT begins. Here, practical directed graph clustering methodologies such as connectivity, fuzzy clustering, and k-means methods may be applied, and below, the connectivity methodology will be described as an example.

를 경험하는 WT로 정의된다. 리드 WT 방향 그래프 연결 정보는 동일한 서브셋에 클러스터링된 터빈의 인접성(adjacency)을 결정한다. 특히, 희소 통신 아키텍처는 분산형(distributed) 또는 비중앙집중형(decentralized)이다(도 2 참조). 분산형(distributed) 통신 아키텍처(빨간 점선 화살표)에는 공유 터빈이 있는 반면, 비중앙집중형(decentralized) 통신 아키텍처에는 서브 셋들이 서로 독립적이며 공유 터빈이 없다. 모든 터빈에 대해 통신 이웃을 결정하기 위하여 웨이크 방향 그래프

를 구성하고 웨이크 계수 행렬(weight coefficient matrix)

를 계산한다. 도 3을 참조하여 희소 통신 아키텍처를 구축하기 위한 프로세스를 설명한다. For a given wind condition, the subset lead WT is the free-flow velocity

is defined as a WT that experiences The lead WT directed graph connection information determines the adjacency of turbines clustered in the same subset. In particular, the sparse communication architecture is either distributed or decentralized (see Fig. 2). In a distributed communication architecture (red dashed arrow) there is a shared turbine, whereas in a decentralized communication architecture the subsets are independent of each other and there is no shared turbine. Wake direction graph to determine communication neighbors for all turbines

and construct a wake coefficient matrix (weight coefficient matrix)

to calculate A process for building a sparse communication architecture is described with reference to FIG. 3 .

, 풍속 U, WT간 거리 X, WT의 반경 D 등 관련 파라미터를 모두 수집한다.Step S310: Coordinates of all WTs (X, Y), wind direction

, wind speed U, distance X between WTs, and radius D of WT are all collected.

Step S320: Transform the WT coordinates using Equation (1).

,

, WT 간 거리 x를 계산한다.Step S330:

,

, compute the distance x between WTs.

를 생성하고, 웨이크 방향 그래프

를 생성한다. 이후, 모든 웨이크 가중치에 대한 평균 값

를 계산한다. Step S340: Wake weighting coefficient matrix based on equations (6) and (7)

and create a wake-directed graph

to create Afterwards, the average value for all wake weights

to calculate

를 정의한다. 이후, WT 서브 셋의 통신 이웃을 줄이기 위해 웨이크 방향 그래프를 가지치기(

)한다. Step S350: select a value of k, and then a threshold value for pruning

to define Then, pruning the wake direction graph to reduce the communication neighbors of the WT subset (

)do.

Step S360: Define the lead WT of each subset, and divide the subset Ni through the connection of the directed graph network topology.

Step S370: Check whether there is a shared WT Si among the subset Ni, to determine whether it is a decentralized communication architecture or a distributed communication architecture.

Step S380: If the shared WT belongs to several subsets, the subset with the maximum weight is maintained, and the distributed type is converted into a decoupled and decentralized communication architecture.

3.3. Example of a sparse communication architecture

를 경험하는 WT로 정의된다. 여기에서, 업스트림 WT는 다운스트림 WT에 영향을 미친다. 어두운 색은 WT들 사이의 강한 상호작용을 나타내며, 밝은 색은 약한 상호작용을 나타낸다.Hereinafter, a method of constructing a sparse communication architecture will be described with specific examples. As shown in FIG. 4 , a WF including 9 WTs having a layout of 3×3 is assumed. The turbine coordinates can be transformed from downstream coordinates to cross-stream coordinates (Fig. 4(a) downstream coordinates, Fig. 4(b) cross-stream coordinates) based on equation (1), through which the lead turbine can be easily determined have. Lead WT is the free stream rate

is defined as a WT that experiences Here, the upstream WT affects the downstream WT. Dark colors indicate strong interactions between WTs, and light colors indicate weak interactions.

와 계수 행렬

에 의해 설명될 수 있다. 웨이크 방향 그래프

는 리드 WT와 이를 따르는 WT들 간 웨이크 인터렉션의 네트워크 토폴로지를 나타낸다. 웨이크 방향 그래프에서 빨간색 점은 WT를 나타내며, 파란색 선과 화살표는 웨이크 흐름과 방향을 나타내고, 라인에 표현된 숫자는 웨이크 가중 계수를 나타낸다. 작은 웨이크 가중 계수는 약한 강도(영향 수준)을 의미한다. 특히, 방향 그래프의 왼쪽 첫번째 열(column)은 리드 WT를 나타낸다. Wake distribution is a wake-direction graph

and coefficient matrix

can be explained by wake direction graph

denotes a network topology of a wake interaction between a lead WT and WTs following it. In the wake direction graph, red dots indicate WT, blue lines and arrows indicate wake flow and direction, and numbers expressed in lines indicate wake weighting coefficients. A small wake weighting factor indicates a weaker intensity (impact level). In particular, the first column on the left of the directed graph represents the lead WT.

5A and 5C show an original wake direction graph and a weighting coefficient matrix. In the original directed graph, the lead WTs are T1, T2, T3, and the subset of adjacent WTs is N1 = {1,4,5,7,8,9}, N2 = {2,5,6,8,9}, N3 = {3,6,9}. In this wind condition, the shared WT numbers for the N1, N2, N3 subsets are S1={5,8,9}, S2={5,6,8,9}, S3={6,9}. As for the WT subset and the shared set, the WT subset and the shared set can be easily defined as shown in Table 1 below through a wake direction graph and a weighting coefficient matrix.

[Table 1]

가지치기법을 통해 일정 값 이하의 웨이크 가중 계수를 제거함으로써 웨이크 방향 그래프의 크기를 축소시킬 수 있다. 보다 구체적으로, 웨이크 가중 계수

가 정의된 임계값

보다 작으면 가중 계수는 0으로 설정된다. (

)When an intrinsic wake-directed graph is sparse, its size is reduced while the underlying architecture of the graph is preserved. Adaptive k-threshold

The size of the wake direction graph can be reduced by removing a wake weighting coefficient less than a predetermined value through the pruning technique. More specifically, the wake weighting factor

is a defined threshold

If less than, the weighting factor is set to zero. (

)

Figures 5c and 5d show the pruning results for Figures 5a and 5b. Lead WTs are T1, T2, T3, T4, T5, T6, T7, and a subset of adjacent WTs is N1 = {1, 8, }, N2={2, 9}, N3={3}, N4={5}, N5={5}, N6={6}, N7={7}. It can be confirmed that there is no shared WT under the corresponding wind conditions, and summarized in Table 2 below.

[Table 2]

4. Coordination Optimization Strategies

When the WF sparse communication architecture is built, the coordination control optimization strategy (algorithm) must be matched with it. As described above, the communication architecture of the WF can be classified into a centralized type, a non-centralized type, and a distributed type.

4.1. Centralized optimization WF Control

Based on the FLORIS wake model described in 2 above, the centralized power optimization function for the entire WF can be expressed as follows.

[Equation (8)]

이고,

와

는 최소 및 최대 요각 범위를 의미하며,

과

는 최소 및 최대 축 계수 범위를 의미하며,

는 최소 전력을,

는 정격 전력을 의미하고, Pj는 WTj에 대한 전력을 의미한다. where f(x) represents the sum of the powers for all N WTs in WF,

ego,

Wow

means the minimum and maximum yaw angle range,

class

means the minimum and maximum axis counting ranges,

is the minimum power,

denotes the rated power, and Pj denotes the power for WTj.

가 주어졌을 때, 함수 값

, 그래디언트

, 헤시안 행렬(Hessian matrix)

을 계산하기 위해 반복이 진행되며, 식(8)은 다음과 같이 SQP 하위 문제(sub-problem)를 정의한다. The WF power function f(x) is a nonlinear programming problem (NLP). Here, the SQP algorithm can be selected in consideration of simple implementation and practical efficiency, and the sub-optimum of Equation (8) can be found using the SQP algorithm. The solution for equation (8) in the kth iteration

Given a function value,

, gradient

, Hessian matrix

Iteration is performed to calculate , and Equation (8) defines the SQP sub-problem as follows.

[Equation (9)]

Here, the SQP may be solved using CVX, which is convex programming software. The SQP algorithm for WF optimization can be summarized as Algorithm 1 below.

[Algorithm 1]

로 정의하고, 우선, 실현 가능한 상승 방향(a feasible ascent direction)

가 계산된다. 이후, 상승 방향

을 이용하여 목적 함수

가 새롭게 반복된다. 여기에서,

는 스텝 크기(step size)이다. 일반적으로, 이러한 증가는

와 같이 반복되며, 여기에서, v는 증가 파라미터(increased parameter)를 의미한다. 그래디언트

가 충분히 작을 때(예 :

) 수렴이 가정될 수 있다. v와

의 일반적인 값은 각각

과

이다. QP's solution

and, first of all, a feasible ascent direction

is calculated Afterwards, the upward direction

objective function using

is repeated anew. From here,

is the step size. In general, this increase

is repeated, where v means an increased parameter. gradient

When is small enough (eg:

) convergence can be assumed. v and

The typical values of each are

class

to be.

Especially in large-scale OWF, the control parameters of WF increase exponentially as the turbine increases. The centralized control method is associated with high communication cost and low computational performance. Therefore, sparse communication architectures and computational efficiency algorithms are needed to achieve speed-up optimization and control.

It should be noted that for small OWFs, the centralized optimization algorithm is suitable in terms of time and computational efficiency, and can adapt to changing atmospheric conditions. However, when the scale of WF is increased, a decentralized or distributed algorithm with high computational efficiency may be suitable for performing optimization and control.

4. 2. Distributed power optimization strategy

를 정의(여기서, xn은 각 서브셋 n의 모든 WT에 대한 요각(yaw angles) 및 축 계수(and axial factors)임)하면, OWF에 대한 전력 함수는 다음과 같다.In a distributed sparse communication architecture, shared nodes exist among subsets. In the distributed power function, constraints on shared nodes must be added, and as a result, the final value of shared nodes must reach consensus and convergence. The ADMM algorithm can be used to solve the distributed optimization problem. To express a function concisely, a vector

, where xn is the yaw angles and axial factors for all WTs in each subset n, the power function for OWF is

[Equation (10)]

은 각 서브셋의 전력 함수, Ns는 서브셋의 수, xs, zs는 공유 WT(s = 1 ~ Ns)이며, z는 x의 복사본으로서 공유 노드가 동일한 값으로 수렴되도록 할 수 있다. 상기 3. 에서 설명한 바와 같이 WT 서브셋과 WF의 공유 세트를 결정한다. 상기 3. 3.의 적용 예(표 1)을 참조하면, 전체 WF 전력 함수는 아래와 같이 표현될 수 있다.here

is the power function of each subset, Ns is the number of subsets, xs, zs is the shared WT (s = 1 to Ns), and z is a copy of x so that the shared nodes can converge to the same value. As described in 3. above, a shared set of WT subsets and WFs is determined. Referring to the application example of 3. 3. (Table 1) above, the entire WF power function can be expressed as follows.

Here, each of the constraints on the shared nodes of f1(x1), f2(x2), and f3(x3) is as follows.

In Equation (10), the distributed optimization problem can be solved by minimizing the augmented Lagrangian using ADMM iteratively.

[Equation (11)]

를 정의하면, 식 (11)의 유도를 통해 다음 식을 얻을 수 있다.scaled double variable

is defined, the following equation can be obtained through the derivation of equation (11).

[Equation (12)]

의 최소화, x, y, u는 개별적으로 반복 업데이트, k은 반복 횟수

minimization of , where x, y, u are individually iterative updates, k is the number of iterations

[Equation (13)]

[Equation (14)]

[Equation (15)]

In the following normalized ADMM iteration formula, fn(xn) can be separated and executed in parallel. After that, the parameter iteration process of the nth subset is as follows.

[Equation (16)]

x update

[Equation (17)]

z update

[Equation (18)]

u update

및 이중 잔차(dual residuals)

를 기반으로 정의된다(

및

는 임계값임). 이 설정은 공유 노드 간의 차이가 0이 되도록 동기를 부여한다. 공유된 WT는 요각과 축 계수가 같아야 하며, 패널티 파라미터

는 공유된 WT 간의 차이를 가중시키는데 사용될 수 있다.

,

및

는 표준 값으로서 각각 0.5,

,

로 설정될 수 있다. 식 (16)과 식 (17)은 비콘벡스(nonconvex) 함수로서, 솔루션

와

을 얻기 위해 상기 알고리즘 1과 유사한 반복 알고리즘이 사용될 수 있다. 비콘벡스 OWF 함수를 위한 SQP를 이용한 분산형 알고리즘은 아래와 같다. Here, x-updates may proceed in parallel. The iterative stopping condition is the primal residuals.

and dual residuals

is defined based on (

and

is the threshold). This setting motivates the difference between shared nodes to be zero. Shared WTs must have the same yaw and axis coefficients, and penalty parameters

can be used to weight differences between shared WTs.

,

and

are the standard values of 0.5, respectively,

,

can be set to Equations (16) and (17) are nonconvex functions,

Wow

An iterative algorithm similar to Algorithm 1 above can be used to obtain . The distributed algorithm using SQP for the beaconvex OWF function is as follows.

[Algorithm 2]

4. 3. Decentralized power optimization strategy

For a decentralized sparse communication architecture, the number of WTs in a subset is small, and there are no shared nodes (WTs) between each subset. Decentralized optimization algorithms are very important for online optimization. Here, all subsets respectively maximize the power function of the non-convex OWF. This function can be expressed as follows.

[Equation (19)]

은 n번째 서브셋의 모든 WT의 요각,

은 n번째 서브셋의 모든 WT의 축 계수(axial factors)이며, 나머지는 식 (8)과 동일하다. 각 WT 서브셋의 수는 WF에 있는 WT 수의 일부에 불과하다. 상기 3. 3.의 적용 예(표 2)를 참조하면, 전체 WF 전력 함수는 아래와 같이 표현될 수 있다. where n is the number of subsets of WF, Ns is the total number of subsets,

is the yaw angle of all WTs in the nth subset,

is the axial factors of all WTs in the nth subset, and the rest are the same as in Eq. (8). The number of each WT subset is only a fraction of the number of WTs in the WF. Referring to the application example of 3. 3. (Table 2) above, the entire WF power function can be expressed as follows.

Here, each subset can independently solve its own optimization problem, and since the optimization process by each subset can be performed in parallel, there is an advantage in that it is possible to minimize the time required to complete the optimization process of a large-scale WF. As with Algorithm 1, the SQP algorithm can be used to optimize each OWF subset function. The decentralized algorithm using SQP for the non-convex OWF function is as follows.

[Algorithm 3]

5. Case Study

)으로 시뮬레이션되었다. 중앙집중형, 비중앙집중형, 분산형 제어기는 CVX로 구현되었다. 이 시뮬레이션에서, 요각

초기값은 0이고 범위는 [-30, 30]으로 설정되었으며, 축 계수

초기 값은 1/3이고 범위는

으로 설정되었다In order to prove the control method based on the scalable wake direction graph according to the present invention, a test was performed on a WF of 150 MW power consisting of 30 NREL-5MW turbines (see Fig. 6a, the longitudinal distance between each turbine is 200 m, , the horizontal distance is 600 m). Wind fields and wake effects can be achieved using the FLORIS platform to achieve variable wind directions and fixed wind speeds (

) was simulated. Centralized, non-centralized, and distributed controllers are implemented with CVX. In this simulation, the yaw angle

The initial value is 0, the range is set to [-30, 30], and the axis factor

The initial value is 1/3 and the range is

was set to

5.1. Scalable Wake Direction Graph in Different Wind Directions

이 0도, 45도 및 175도에 대한 시뮬레이션 결과를 설명한다. As described in 3. above, the sparse communication architecture strategy according to the present invention relies on a scalable wake direction graph with different wind directions. Below, the wind direction

Simulation results for these 0 degrees, 45 degrees and 175 degrees will be described.

이 0도일 때의 웨이크 방향 그래프(

)로, 6a는 웨이크 필드, 6b는 분산형 웨이크 방향 그래프(

), 6c는 원래(original)의 웨이크 방향 그래프(

) 및 6d는 비중앙집중형 웨이크 방향 그래프(

)이다. 6 is a wind direction

Wake direction graph when this is 0 degree (

), where 6a is the wake field, and 6b is the distributed wake direction graph (

), 6c is the original wake direction graph (

) and 6d are decentralized wake-directed graphs (

)to be.

이 45도일 때의 웨이크 방향 그래프(

)로, 7a는 웨이크 필드, 7b는 분산형 웨이크 방향 그래프(

), 7c는 원래(original)의 웨이크 방향 그래프(

) 및 7d는 비중앙집중형 웨이크 방향 그래프(

)이다. 7 is a wind direction

Wake direction graph when this is 45 degrees (

), where 7a is the wake field, and 7b is the scatter-directed wake graph (

), 7c is the original wake direction graph (

) and 7d are decentralized wake-directed graphs (

)to be.

이 175도일 때의 웨이크 방향 그래프(

)로, 8a는 웨이크 필드, 8b는 분산형 웨이크 방향 그래프(

), 8c는 원래(original)의 웨이크 방향 그래프(

) 및 8d는 비중앙집중형 웨이크 방향 그래프(

)이다. 8 is a wind direction

Wake direction graph when this is 175 degrees (

), where 8a is the wake field, and 8b is a scatter-directed wake graph (

), 8c is the original wake direction graph (

) and 8d are decentralized wake-directed graphs (

)to be.

A network topology of the wake direction graph may be established in various ways. Referring to the wake direction graphs of FIGS. 6 to 8 , it can be seen that various changes can be made according to the wind direction. For example, when the wind direction is 0 degrees, the wake connection relationship of the wind field is complicated, and correspondingly, the network topology is also complicated as shown in FIGS. 6A and 6C . On the other hand, when the wind direction is 45 degrees, the wake connection relationship and network topology are simple as shown in FIGS. 7A and 7B . This indirectly proves that the wake direction graph according to the present invention can account for the changing wake field.

의 임계값에 의해 정량적으로 조정됨으로써 확장 가능하다. 도 6 내지 8을 참조하면 하이퍼 파라미터 k가 조정됨에 따라 웨이크 방향 그래프 네트워크 토폴로지가 확장될 수 있음을 확인할 수 있다. 이러한 방법은 그래프의 적응형 임계값 가지치기에 의해 특정 꼭짓점 또는 엣지를 직접 제거할 수 있으며, 이에 원래의 그래프 토폴로지를 유지하여 추론을 위한 컴팩트하고 효율적인 그래프를 얻을 수 있다.Also, the wake directed graph network topology is

It is scalable by being quantitatively adjusted by the threshold of . 6 to 8 , it can be seen that the wake-directed graph network topology can be expanded as the hyper parameter k is adjusted. This method can directly remove specific vertices or edges by adaptive threshold pruning of the graph, thereby maintaining the original graph topology to obtain a compact and efficient graph for inference.

The clustered subset and shared WT can be easily identified through network topology analysis of distributed and decentralized wake-directed graphs, and then the construction process described in 3.3.

[Table 3]

[Table 4]

Tables 3 and 4 show the results of clustering in a wind direction of 175 degrees, Table 3 is a distributed wake direction graph, and Table 4 is a non-centralized wake direction graph.

는 0.11, 0.0367 및 0.08이다. 만약 세 개의 방향 그래프가 동일한 고정 임계값(예 : 0.08)을 사용하는 경우, 평균이 0.0367인 방향 그래프의 기본 네트워크 아키텍처가 파손될 수 있다. 이에, 웨이크 방향 그래프마다 상이한 임계값이 적응적으로 적용될 수 있다. 또한, 방향 그래프의 하이퍼 파라미터 k와 희소성(sparsity)은 역비례 관계이다.The present invention applies the adaptive threshold method, which defines the average value of the weighting coefficients of the directed graph as a threshold value and adjusts the threshold value with the hyperparameter k. In fact, the wake-directed graph and wake weights show changes in different wind conditions, and when a fixed threshold is used for pruning for all wake-directed graphs, the wake-directed graph network topology can change dramatically. For example, the average threshold in each of the three wind directions

are 0.11, 0.0367 and 0.08. If all three directed graphs use the same fixed threshold (eg 0.08), the underlying network architecture of the directed graph with an average of 0.0367 may be broken. Accordingly, different threshold values may be adaptively applied to each wake direction graph. Also, the hyperparameter k of the directed graph and the sparsity are inversely proportional to each other.

5.2. Optimization of coordination in different wind directions

When the WF sparse communication architecture is built, the OWF power optimization problem is solved by using a corresponding distributed algorithm (Equation (10)) or a decentralized algorithm (Equation (19)). Hereinafter, simulation results at 0, 15, 30, 45, 60, and 175 degree wind directions will be described. To verify the performance of the algorithm according to the present invention, a comparison between a greedy algorithm and a centralized algorithm is performed. Each WT power distribution is shown in Figure 6a. Here, the WF power efficiency may be defined as follows.

[Equation (20)]

는 WT가

풍속에서 웨이크 간섭이 존재하지 않는다고 가정했을 때 생성할 수 있는 최대 전력이다. 이러한 표준화로 인해, WF 전력 효율

는 자유 흐름 풍속

에 의존하지 않는다. 특히, 단일 터빈의 최대 전력은

에서

이며, 30개 터빈의 최대 전력은

이다. here

is the WT

This is the maximum power that can be generated at wind speed, assuming no wake interference exists. Due to this standardization, WF power efficiency

is the free flow wind speed

does not depend on In particular, the maximum power of a single turbine is

at

and the maximum power of 30 turbines is

to be.

9 shows the wind power distribution in each wind direction ((a) to (f) show the power distribution at 0, 15, 30, 45, 60, and 175 degrees, respectively). Here, a change in wind direction induces a change in power and optimization results in each unit. When the greedy control method is applied, the upstream direction unit generates the maximum power of a single unit, while the power obtained by the downstream direction unit is greatly reduced. When the WT subset follows a coordinated control scheme, the power of the upstream turbine is reduced while the power of the downstream turbine is greatly increased, increasing the overall power of the OWF.

)을 생산함을 알 수 있다. 반면, 다운스트림 터빈은 업스트림 터빈에 의한 웨이크 영향으로 낮은 전력을 생산함(T9, T16, T26, T27는 각각 0.318, 0.318, 0.41, 0.318의 전력 효율을 나타냄)을 알 수 있으며, 전체 전력 효율은 0.58을 나타낸다. 10 shows the WT power energy efficiency results for four control methods in a wind direction of 30 degrees ((A) is greedy control, (B) is centralized control, (C) is non-centralized control, ( D) is a decentralized control). For greedy control, the maximum power (

) can be seen to produce On the other hand, it can be seen that the downstream turbine produces low power due to the wake effect by the upstream turbine (T9, T16, T26, and T27 show power efficiency of 0.318, 0.318, 0.41, and 0.318, respectively), and the overall power efficiency is 0.58.

In the case of centralized, decentralized and distributed optimization methods, the downstream turbine produces higher power compared to the greedy control method. Referring to FIG. 10(B), when the centralized method is used, the power efficiency of the lead turbine T2 is reduced, but the power efficiency of the downstream turbine is increased (T9, T16, T26, and T27 are 0.424, 0.424, and 0.614, respectively) , representing a power efficiency of 0.534), and the overall power efficiency is 0.6558. Referring to FIG. 10(C), it can be seen that the power efficiency of the downstream turbines T9, T16, T26, and T27 is 0.425, 0.425, 0.616, and 0.534, respectively, when using the decentralized method, The overall power efficiency is 0.6555. Referring to FIG. 10(D), it can be seen that the power efficiencies of the downstream turbines T9, T16, T26, and T27 are 0.418, 0.417, 0.616, and 0.528, respectively, when using the distributed method, and the total power The efficiency shows 0.6557. In the end, it can be seen that the three regulated control modes improve the overall power efficiency of the WF.

Table 5 below shows the results for the power efficiency (PE) and calculation time (CT) of the WF in each wind direction. PE has no unit, and CT has a unit of seconds (sec).

[Table 5]

By analyzing Table 5, the following conclusions can be drawn.

(1) Power efficiency (PE) and calculation time (CT) appear differently depending on the wind direction and communication architecture. This is because the wake direction graph and the weighting factor are different depending on the wind direction, and the change in the communication architecture has a significant effect on the PE and CT results. Let us consider the case where the wind direction is 0 degrees and 45 degrees, respectively. First, in the case of 0 degree, the interaction between turbines appears, and the distributed and decentralized sparse directed graphs have a complex communication architecture. On the other hand, in the case of 45 degrees, weak interaction between turbines appears, and the distributed and non-centralized sparse directed graphs have a simple communication architecture. It can be inferred that the more complex the communication architecture, the longer the corresponding CT.

(2) Centralized, non-centralized and distributed optimization algorithms according to the communication architecture can monotonically improve the total PE of WF. In particular, the PE of the greedy algorithm is the lowest, followed by the decentralized algorithm, and the efficiency of the distributed algorithm is almost the same as that of the centralized control mode. However, a non-centralized control method in which power output efficiency is not significantly lowered compared to centralized and distributed methods may be most appropriate from the viewpoint of real-time control.

(3) The non-centralized method based on the defined sparse directional graph communication structure can reduce the computational complexity without significant loss in terms of accuracy, and the computation time is less than that of the centralized and distributed methods. For example, when the wind direction is 0 degrees, CT is reduced compared to centralized and distributed algorithms, and there is no significant difference in power efficiency.

However, there is always a trade-off between PE and CT. When only CT is considered, greedy control is appropriate, but when only PE is considered, centralized control may be appropriate. Decentralized and decentralized control are intermediate between the two methods, and which method to choose may depend on actual needs.

6. Power adjustment optimization device according to the present invention

Hereinafter, an apparatus for optimizing power adjustment that performs the above-described method for optimizing power adjustment will be described with reference to FIG. 11 .

The apparatus for optimizing power adjustment according to the present invention is a computing device that performs the above-described method for optimizing power adjustment, and may include a communication architecture determining unit 100 and a power adjustment optimization unit 200 . Here, the configurations of the power adjustment optimization apparatus may be implemented by being included in a single device (eg, a central control unit), and implemented by being distributed to a plurality of devices (eg, the central control unit and the control unit included in each WT). can be On the other hand, this example is not intended to limit the scope of the present invention, and if the device in which the above-described power adjustment optimization method is implemented, the power adjustment optimization apparatus according to the present invention is not limited to the type, name, number of implemented devices, etc. should be interpreted as

Hereinafter, operations of the communication architecture determining unit 100 and the power adjustment optimization unit 200 will be described, and content overlapping with the previously described content will be omitted.

The communication architecture determining unit 100 determines a communication architecture applied to a plurality of WTs.

The power adjustment optimization unit 200 applies a predefined power adjustment optimization algorithm corresponding to the determined communication architecture to generate a power adjustment optimization solution capable of adjusting the yaw angle and axis coefficient of each WT.

That is, in performing the power adjustment optimization method described in 1. to 5., the communication architecture determining unit 100 determines a communication architecture suitable for the WT, and the power adjustment optimization unit 200 performs the power adjustment optimization strategy. It is a configuration that performs the process of building a .

In one embodiment, the communication architecture determining unit 100, when the number of WTs included in the WF is less than or equal to a predefined number, the communication architecture applied to the plurality of WTs may be determined to be centralized. Here, the power adjustment optimization unit 200 may generate a power adjustment optimization solution by applying the above-described centralized optimization algorithm.

In one embodiment, the communication architecture determining unit 100 collects parameters (coordinates of the WT, wind direction, wind speed, radius of the WT, etc.) for a plurality of WTs, which are related to the output power of the WF. Thereafter, the communication architecture determining unit 100 generates a wake weighting coefficient matrix numerically representing the wake effect acting between a plurality of WTs, and using the WT as a node and the WT as a direction line for the wake effect acting between the WTs. It is possible to create a wake direction graph represented by .

Here, the communication architecture determiner 100 may perform a pruning process on the wake weighting coefficient matrix and the wake direction graph by using a threshold defined based on the wake weighting coefficient. In an embodiment, the communication architecture determiner 100 may calculate an average value for all wake weighting coefficients, and define the calculated average value as the threshold value. Here, the communication architecture determiner 100 may adjust the defined threshold based on a hyper parameter that controls the sparse degree of the wake direction graph. The communication architecture determiner 100 may perform a pruning process on the wake weighting coefficient matrix and the wake direction graph based on the finally determined threshold.

Thereafter, the communication architecture determining unit 100 clusters a plurality of WTs based on the wake weighting coefficient matrix and the wake direction graph on which the pruning process is completed and classifies them into a plurality of subsets, and based on the subset classification result, the plurality of WTs It is possible to determine the sparse communication architecture applied to

As described above, the sparse communication architecture may be classified into a decentralized communication architecture and a distributed communication architecture, and the communication architecture determining unit 100 determines that the shared WT between the classified subsets is By judging whether or not the shared WT exists, it may be determined as a non-centralized communication architecture if the shared WT does not exist, and a distributed communication architecture may be determined if the shared WT exists.

When the applied sparse communication architecture is determined to be a non-centralized communication architecture, the power adjustment optimization unit 200 may apply the above-described non-centralized optimization algorithm to generate a power adjustment optimization solution. If the applied sparse communication architecture is determined as the distributed communication architecture, the power adjustment optimization unit 200 may apply the above-described distributed optimization algorithm to generate a power adjustment optimization solution.

The power adjustment optimization method according to the present invention described above may be implemented as computer-readable codes on a computer-readable recording medium. The computer-readable recording medium includes all types of recording devices in which data readable by a computer system is stored. Examples of computer-readable recording media include ROM, RAM, CD-ROM, magnetic tape, floppy disk, and optical data storage device. In addition, the computer-readable recording medium is distributed in a computer system connected to a network, so that the computer-readable code can be stored and executed in a distributed manner.

The above-described preferred embodiments of the present invention have been disclosed for purposes of illustration, and various modifications, changes, and additions may be made by those skilled in the art with respect to the present invention within the spirit and scope of the present invention, and such modifications, changes and Additions are to be considered as falling within the scope of the following claims.

100: communication architecture decision unit
200: power adjustment optimization unit

Claims (12)

A method for optimizing power adjustment of a wind power farm (WF) comprising a plurality of wind turbines (WT), the method comprising:
collecting a parameter for the plurality of WTs that is associated with an output power of the wind farm;
generating a wake weighting coefficient matrix numerically representing a wake effect acting between the plurality of WTs, and a wake direction graph representing a wake effect acting between the WTs as a directional line using the WT as a node;
performing a pruning process on the wake weighting coefficient matrix and the wake direction graph using a threshold defined based on a wake weighting coefficient;
classifying the plurality of WTs into a plurality of subsets by clustering the plurality of WTs based on a wake weighting coefficient matrix and a wake direction graph on which the pruning process is completed;
determining a sparse communication architecture applied to the plurality of WTs based on the subset classification result; and
generating a power adjustment optimization solution capable of adjusting the yaw angle and axis coefficient of each WT by applying a predefined power adjustment optimization algorithm corresponding to the determined sparse communication architecture; Way.
The method of claim 1, wherein performing the pruning process comprises:
Calculating an average value for all wake weighting coefficients, and defining the calculated average value as the threshold value.
3. The method of claim 2, wherein performing the pruning process comprises:
Adjusting the defined threshold based on a hyper parameter for controlling the sparse degree of the wake direction graph; The method of optimizing power adjustment of the wind farm further comprising a.
The method of claim 1, wherein the sparse communication architecture comprises:
A method for optimizing power coordination of a wind farm, comprising: a decentralized communication architecture and a distributed communication architecture.
5. The method of claim 4, wherein determining the sparse communication architecture comprises:
determining whether there is a shared WT among the classified subsets; and
If the shared WT does not exist, determining as a non-centralized communication architecture, and if the shared WT exists, determining as a distributed communication architecture.
6. The method of claim 5, wherein generating the power regulation optimization solution comprises:
When the applied sparse communication architecture is determined to be a non-centralized communication architecture, a predefined decentralized optimization algorithm is applied to generate the power adjustment optimization solution;
When the applied sparse communication architecture is determined to be a distributed communication architecture, a method for optimizing power regulation of a wind farm, characterized in that the power regulation optimization solution is generated by applying a predefined distributed optimization algorithm.
An apparatus for optimizing power adjustment of a wind farm for maximum power output of a wind farm (WF) comprising a plurality of wind turbines (WT), the apparatus comprising:
a communication architecture determining unit that determines a communication architecture applied to the plurality of WTs; and
A power adjustment optimization unit for generating a power adjustment optimization solution capable of adjusting the yaw angle and axis coefficient of each WT by applying a predefined power adjustment optimization algorithm corresponding to the determined communication architecture;
The communication architecture determining unit,
collect parameters for the plurality of WTs, which are associated with output power of the wind farm;
A wake weighting coefficient matrix numerically representing a wake effect acting between the plurality of WTs, and a wake direction graph representing the wake effect acting between the WTs as a directional line using the WT as a node,
performing a pruning process on the wake weighting coefficient matrix and the wake direction graph using a threshold defined based on a wake weighting coefficient;
Classifying the plurality of WTs into a plurality of subsets by clustering the plurality of WTs based on a wake weighting coefficient matrix and a wake direction graph on which the pruning process is completed,
determining a sparse communication architecture applied to the plurality of WTs based on the subset classification result,
Power regulation optimization device for wind farms.
The method of claim 7, wherein the communication architecture determining unit,
An apparatus for optimizing power adjustment of a wind farm, characterized in that calculating an average value for all wake weighting coefficients, and defining the calculated average value as the threshold value.
The method of claim 8, wherein the communication architecture determining unit,
Power adjustment optimization apparatus of a wind farm, characterized in that adjusting the defined threshold based on a hyper parameter that controls the sparse degree of the wake direction graph.
8. The method of claim 7, wherein the sparse communication architecture comprises:
A device for optimizing power regulation of a wind farm, characterized in that it includes a decentralized communication architecture and a distributed communication architecture.
The method of claim 10, wherein the communication architecture determining unit,
It is determined whether a shared WT exists between the classified subsets, and if the shared WT does not exist, a decentralized communication architecture is determined, and if the shared WT exists, a distributed communication architecture is determined. Power regulation optimization device for wind farms.
The method of claim 11, wherein the power adjustment optimization unit,
When the applied sparse communication architecture is determined to be a non-centralized communication architecture, a predefined decentralized optimization algorithm is applied to generate the power adjustment optimization solution, and the applied sparse communication architecture is a distributed communication architecture. When it is determined, the apparatus for optimizing power adjustment of a wind farm, characterized in that by applying a predefined distributed optimization algorithm to generate the power adjustment optimization solution.

Images