A distributed robust optimization scheduling method for offshore wind power clusters considering spatiotemporal correlation

By constructing a fuzzy set and decomposition optimization model of spatiotemporal correlation, the problem of unconsidered space-time correlation in offshore wind power optimization scheduling is solved, computing efficiency and economy are improved, and the safe and economic operation of wind power clusters are achieved.

CN114825347BInactive Publication Date: 2024-11-12SOUTH CHINA UNIV OF TECH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202210304741.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-03-22
Publication Date
2024-11-12
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The prior art fails to effectively consider the collecting network structure and current distribution in the wind farm in the offshore wind power optimization scheduling, and the distribution robust optimization method fails to make full use of the spatial and temporal correlation, resulting in low computing efficiency and insufficient economicality.

Method used

The fuzzy set of space-time correlation is constructed using the KL divergence metric, and a two-stage distribution robust optimization scheduling model of multi-end VSC-HVDC grid-connected are established. The model is decomposed through the C&CG algorithm and generalized Benders decomposition method to reduce the solution complexity and calculation requirements.

Benefits of technology

It improves computing efficiency, reduces energy consumption, ensures the safe and economic operation of offshore wind power clusters under the harshest wind speed distribution, and optimizes the fan start-stop combination and effective efforts decision-making.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114825347B_ABST
    Figure CN114825347B_ABST
Patent Text Reader

Abstract

The present invention discloses a distributed robust optimization scheduling method for offshore wind power clusters considering spatiotemporal correlation, including: inputting offshore wind farm wind speed data, using KL divergence to measure the distance between the actual joint probability distribution of wind speed and the statistical reference joint probability distribution, and constructing a data-driven fuzzy set considering the spatiotemporal correlation between the wind speeds of multiple wind farms in the same cluster; establishing a two-stage distributed robust optimization scheduling model for offshore wind power clusters connected to the grid via multi-terminal VSC‑HVDC according to the fuzzy set; solving the optimization scheduling model and outputting the result. The present invention reflects the correlation between wind speeds in adjacent time periods of wind farms and the correlation between wind speeds of multiple adjacent wind farms in a cluster by using a joint probability distribution. A novel and efficient distributed robust optimization scheduling solution algorithm is proposed for the established optimization scheduling model. While maintaining accuracy, the algorithm has higher computational efficiency than the C&CG algorithm and meets the requirements of actual scheduling.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of optimization scheduling, and in particular to a distributed robust optimization scheduling method for offshore wind power clusters taking into account temporal and spatial correlation. Background Art

[0002] At present, the main approach of existing technologies for optimizing offshore wind power scheduling is to establish an optimization scheduling model on the grid side and treat the total output of the wind farm cluster as an uncertain variable, without considering the optimization of the internal collection network flow of the wind farm cluster and the optimization of the start-stop combination of the units.

[0003] In dealing with the uncertainty of offshore wind farm output, the main methods currently include stochastic optimization methods, robust optimization methods and distributed robust optimization methods. The distributed robust optimization method combines the advantages of stochastic optimization methods and robust optimization methods, and makes optimal scheduling decisions under the worst probability distribution of uncertain variables. It has been widely used in recent years, but the correlation between uncertain variables is rarely considered, and the correlation between wind speeds in adjacent time periods of wind farms and the correlation between wind speeds of multiple adjacent wind farms in a cluster are not considered at the same time.

[0004] Although existing technologies have begun to optimize the scheduling of power systems containing offshore wind farms, they have not taken into account the collection network structure and tidal distribution within the offshore wind farm, and have not given sufficient consideration to the economic efficiency of the internal operation of the offshore wind farm cluster.

[0005] Although distributionally robust optimization methods have been widely used to deal with uncertainty problems in power systems, most of the current distributionally robust optimization methods do not consider the spatiotemporal correlation between random variables. Only a few methods use the covariance matrix to consider the correlation, but the introduction of the covariance matrix turns the optimization scheduling model into a semi-definite programming model with low solution efficiency; and do not simultaneously consider the correlation between the wind speeds in adjacent time periods of wind farms and the correlation between the wind speeds of multiple adjacent wind farms in a cluster.

[0006] In solving distributed robust optimization models, the column and constraint generation (C&CG) algorithm is the mainstream algorithm. However, during the iterative process of the C&CG algorithm, the scale of the main problem continues to increase, which makes the model solution take a long time, places high requirements on the computer's operating performance, and consumes a lot of energy. Summary of the invention

[0007] In order to solve at least one technical problem existing in the above-mentioned background technology, the present invention provides a distributed robust optimization scheduling method for offshore wind power clusters considering time and space correlation.

[0008] To achieve the above object, the technical solution of the present invention is:

[0009] A distributed robust optimization scheduling method for offshore wind power clusters considering temporal and spatial correlations comprises:

[0010] The wind speed data of offshore wind farms is input, and the distance between the actual joint probability distribution of wind speed and the statistical reference joint probability distribution is measured by KL divergence. A data-driven fuzzy set considering the spatiotemporal correlation between wind speeds of multiple wind farms in the same cluster is constructed.

[0011] A two-stage distributed robust optimization scheduling model for offshore wind power clusters connected to the grid via multi-terminal VSC-HVDC is established based on the fuzzy set. In the first stage, the wind turbine start-stop plan is optimized under the worst wind speed probability distribution to obtain the optimal wind turbine start-stop combination. In the second stage, the active output of the wind turbine is optimized. Under the given wind turbine start-stop state, the worst wind speed probability distribution that maximizes the expected value of the operating cost is obtained.

[0012] The optimization scheduling model is solved and the result is output.

[0013] Compared with the prior art, the present invention has the following beneficial effects:

[0014] (1) A fuzzy set considering the spatiotemporal correlation of wind speed is designed. The KL divergence is used to measure the probability distance between the actual joint distribution of wind speed and the statistical reference joint distribution, and a data-driven fuzzy set considering the correlation between wind speeds in adjacent periods of wind farms and the correlation between wind speeds of multiple adjacent wind farms in a cluster is constructed.

[0015] (2) A two-stage distributionally robust optimal dispatch (DROD) model for offshore wind power clusters connected to the grid via multi-terminal VSC-HVDC was established. This model can optimize the start-stop combination of wind turbines and the active power output under the worst wind speed distribution. The optimization model is converted into a mixed integer second-order cone programming model to reduce the complexity of the model solution.

[0016] (3) An efficient solution algorithm was designed. The C&CG algorithm was used to divide the two-stage robust optimization scheduling model into a main problem and sub-problems for alternating iterative solutions. The generalized Benders decomposition method was used to decompose the main problem into a mixed integer linear programming model containing all 0-1 variables and a continuous second-order cone programming model for alternating iterative solutions to reduce the scale of the main problem and improve computational efficiency, thereby reducing computer performance requirements and reducing energy consumption. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 A flow chart of a distributed robust optimization scheduling method for offshore wind power clusters considering spatiotemporal correlation provided by an embodiment of the present invention;

[0018] Figure 2 It is a two-dimensional wind speed value distribution diagram for adjacent time periods;

[0019] Figure 3 is the two-dimensional wind speed joint frequency histogram of adjacent time periods;

[0020] Figure 4 It is the steady-state equivalent model diagram of VSC;

[0021] Figure 5 It is the active output characteristic curve of the fan;

[0022] Figure 6 Flowchart of the algorithm for solving the optimal scheduling model;

[0023] Figure 7 This is the network topology diagram of the offshore wind power cluster;

[0024] Figure 8 Provide a daily scheduling plan graph for a given cluster;

[0025] Fig. 9 The relative position diagram of wind towers in each wind farm;

[0026] Fig.10 is the relative position diagram of wind turbines in the wind farm;

[0027] Fig.11 It is the number diagram of wind turbines in operation in each wind farm;

[0028] Fig.12 This is the expected value diagram of active power injected into VSC by each wind farm under the worst distribution;

[0029] Fig.13 This is the diagram of the number of wind turbines operating under different optimization models;

[0030] Fig.14 The total active output diagram of the cluster for different models. DETAILED DESCRIPTION

[0031] Example:

[0032] The technical solution of the present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0033] See also Figure 1 As shown, the offshore wind power cluster distributed robust optimization scheduling method considering time and space correlation provided by this embodiment mainly includes the following steps:

[0034] The wind speed data of offshore wind farms is input, and the distance between the actual joint probability distribution of wind speed and the statistical reference joint probability distribution is measured by KL divergence, so as to construct a data-driven fuzzy set that considers the spatiotemporal correlation between the wind speeds of multiple wind farms in the same cluster. In other words, the correlation between the wind speeds of wind farms in adjacent periods and the correlation between the wind speeds of multiple adjacent wind farms in a cluster are reflected by using the joint probability distribution.

[0035] A two-stage distributionally robust optimal dispatch (DROD) model for offshore wind power clusters connected to the grid via multi-terminal VSC-HVDC is established based on the fuzzy sets. In the first stage, the wind turbine start and stop plan is optimized under the worst wind speed probability distribution to obtain the optimal wind turbine start and stop combination. In the second stage, the active output of the wind turbine is optimized. Under given wind turbine start and stop states, the worst wind speed probability distribution that maximizes the expected value of the operating cost is obtained.

[0036] In this way, the start-stop combination of wind turbines and the active output can be optimized under the worst wind speed distribution. The start-stop combination of each wind turbine and the active output plan obtained can ensure the safe and economical operation of the offshore wind power cluster under the worst probability distribution of wind speed.

[0037] The optimization scheduling model is solved and the result is output.

[0038] Specifically, the above probability distribution fuzzy set considering wind speed correlation is constructed as follows:

[0039] Assume that the true probability distribution of the uncertain variable of wind speed in wind farms is discrete. For the wind speeds of multiple wind farms in any adjacent time period, the relevant multi-dimensional joint distribution can be generated by driving the use of historical data. Assume that the vector v t represents the wind speed of M wind farms in time period t, that is, v t =(v 1,t ,...,v m,t ,...,v M,t ), where v m,t is the wind speed of wind farm m in period t; then the M-dimensional discrete joint distribution of wind speeds of M wind farms in period t p(v 1,t ,...,v m,t ,...,v M,t ) can be expressed as The 2M-dimensional discrete joint distribution of wind speeds of M wind farms in time period t and time period t+1 can be expressed as p(v t ,v t+1 ), then the joint probability distribution fuzzy set based on KL divergence distance is constructed as shown in formula (1).

[0040]

[0041] Where p0 represents the corresponding reference joint probability distribution, which can be driven by a large amount of historical data; D KL {p(v t ,v t+1 )||p0(v t ,v t+1 )} represents the joint probability distribution p(v t ,v t+1 ) and p0(v t ,v t+1 ) is the KL divergence distance between (t,t+1),l and p 0(t,t+1),l They are respectively the true distribution p(v t ,v t+1 ) and the reference distribution p0(v t ,v t+1 ) is the probability value of the lth discrete scenario; N (t,t+1) is the joint distribution p(v t ,v t+1 ) is the total number of discrete value scenarios; λ is the KL divergence threshold.

[0042] The discrete value scenarios of high-dimensional joint probability distribution are usually highly sparse. Assume Ω s (t,t+1) is the joint distribution p(v t ,v t+1 ) in the sample space, then Ω s There must be some wind speed discrete value scenarios in (t, t+1), which are extremely low probability events in this space. Eliminating these scenarios from the sample space can greatly reduce the number of value scenarios while considering the correlation. From the non-negativity of probability measure, it can be known that when a value in the low-dimensional marginal probability distribution is known to be an extremely low probability event, all the scenarios corresponding to this value in the joint probability distribution must also be extremely low probability events. Therefore, we can first count the sample spaces of each low-dimensional marginal probability distribution, and then count the sample space Ω of the high-dimensional joint distribution. s (t, t+1) is reduced. The following takes the two-dimensional joint probability distribution as an example to illustrate the construction process of the sample space. First, the wind speed of the wind farm is statistically analyzed using historical data to obtain the fluctuation range of the wind speed [v min ,v max ]; divide the interval into a finite number of small intervals, take the center value of each small interval as the discrete value scenario of a single wind speed, and use Ω v represents the set of all discrete value scenarios; the sample space of the joint probability distribution of wind speed in two adjacent time periods includes two v m,t ,v m,t+1Any combination of all discrete values ​​of wind speed, a value combination constitutes a discrete value scenario. Figure 2 As shown, a discrete scenario of the joint distribution corresponds to a small square in the figure. By counting the frequency of historical samples in each small square, we can get a frequency histogram, such as Figure 3 shown.

[0043] When the number of samples is large enough, the frequency of occurrence of each discrete scene can be taken as probability, and the sample space of the joint probability distribution can be constructed based on the probability of each sample point. Figure 3 The frequency of the discrete values ​​corresponding to the dotted area is 0, that is, the probability is 0, which can be regarded as an impossible event. Only the discrete scenarios corresponding to the red area are selected as the sample space of the joint probability distribution. Assume that the frequency of the sth discrete scenario in the sample space is A s , the total number of historical samples is A, then the statistical reference probability of this scenario is:

[0044] p0{v m,t =v t,s ,v m,t+1 =v t+1,s =A s / A (2)

[0045] According to the above method, the sample space of the wind speed joint probability distribution among multiple wind farms in each period and the wind speed joint probability distribution of each wind farm in every two adjacent periods is statistically screened. M (t),Ω m (t, t+1) represent the joint probability distribution of wind speed of M wind farms in period t The joint probability distribution of wind speed at wind farm m in period t and t+1 is p(v m,t ,v m,t+1 ) The sample space after screening, then the joint probability distribution p(v t ,v t+1 )The elements of the sample space after screening (v 1,t ,...,v m,t ,...,v M,t ,v 1,t+1 ,...,v m,t+1 ,...,v M,t+1 ) must meet the following requirements:

[0046]

[0047]

[0048] If the probability distribution of wind speeds in two adjacent time periods of multiple wind farms is known, p(v t ,v t+1), then the joint probability distribution of wind speeds of multiple adjacent wind farms in time periods t and t+1 is According to the marginal distribution law, we can obtain:

[0049]

[0050]

[0051] In the formula, They are the joint probability distribution of wind speed of multiple wind farms in time periods t and t+1 respectively. The probability value of the sth discrete scenario. t,l,s is the joint probability distribution of two time periods p(v t ,v t+1 ) and the single-period joint probability distribution The correlation coefficient is, when p(v t ,v t+1 ) corresponds to the lth discrete scene in When the sth discrete scenario in A t,l,s Take 1, otherwise take 0; A t+1,l,s is the joint distribution of two periods p(v t ,v t+1 ) and the single period joint distribution The correlation coefficient is the same as A t,l,s Similarly, equations (5)-(6) maintain the joint probability distribution of wind speeds of multiple adjacent wind farms in each period in each high-dimensional joint probability distribution p(v t ,v t+1 ). Formula (1), (5)-(6) constitute the joint probability distribution fuzzy set Ψ considering the spatiotemporal correlation of wind speed. It establishes a fuzzy set based on the KL divergence distance for the joint probability distribution of wind speeds in adjacent time periods of multiple wind farms in the cluster, and can consider the spatiotemporal correlation of wind speeds in actual situations in the fuzzy set.

[0052] Specifically, the DROD model of the above-mentioned multi-terminal VSC-HVDC grid-connected offshore wind power cluster includes:

[0053] (1) Objective function

[0054] The established DROD model for multi-terminal VSC-HVDC grid-connected offshore wind power cluster is a two-stage optimization model: in the first stage, the wind turbine start-stop plan is optimized under the worst wind speed probability distribution to obtain the optimal wind turbine start-stop combination. The start-stop variable in this stage is represented by x; in the second stage, the active output of the wind turbine is optimized. Under the given wind turbine start-stop state, the worst wind speed probability distribution that maximizes the expected value of the operating cost is obtained. The variables in the second stage are represented by p and y. At this time, the output y of the wind turbine can be adjusted according to the uncertain fluctuation of wind speed p to obtain the optimal wind turbine output decision. The objective function of the DROD model is as follows:

[0055]

[0056]

[0057] In the formula, E(·) represents the expected operation; T is the total number of scheduling periods, which is 24 in this paper; N g is the total number of fans in the cluster; a g ,b g ,c g ,d g They represent the startup cost coefficient, shutdown cost coefficient, operation cost coefficient and idle cost coefficient of fan g respectively; I g,t represents the operating state variable of fan g in time period t, and "1 / 0" represents the operation / stop of the fan; z g,t is the start variable of fan g in time period t, “1” indicates that the fan is started, otherwise “0” is taken; u g,t is the shutdown variable of fan g in time period t, taking "1" to indicate that the fan is shut down, otherwise taking "0";

[0058] Considering that the probability distribution of wind speed is discrete, the expected calculation formula can be obtained from the joint probability distribution of fuzzy sets:

[0059]

[0060]

[0061]

[0062]

[0063] In the formula, and The joint probability distribution of multiple wind farms in time periods 1 and T are and The sth discrete probability value of is obtained by equations (5)-(6), N1 and N T They are and The total number of scenes in the sample space, and N t yes The total number of scenes in the sample space. Q(y t,s ,v t ) including p(v t ) corresponds to the generation deviation penalty cost and network loss cost of the sth discrete scenario, C1 and C2 are the generation deviation penalty cost coefficient and network loss cost coefficient respectively; P ∑,t,s is the net active power output of the wind power cluster in scenario s during period t; P d,t It is the active power dispatching plan value sent by the power grid dispatching center to the wind power cluster; Ω E is the set of branches in the cluster; r ij is the resistance value of branch ij; Q(y is the square value of the current flowing through line ij in scenario s during time period t. t,s ,v t ) contains absolute value calculations, which are difficult to solve directly due to the characteristics of piecewise functions. In order to reduce the computational difficulty of solving the model, equations (13) to (15) are used to equivalently replace equation (12) containing absolute value calculations.

[0064]

[0065]

[0066]

[0067] (2) Constraints

[0068] 1) AC collection network operation constraints

[0069] The AC collection network of an offshore wind farm is usually a radial network, which can be described by a branch power flow model as follows:

[0070]

[0071]

[0072]

[0073]

[0074] In the formula, δ(j) / π(j) represents the set of nodes whose parent node / child node is j; x ij is the reactance value of line ij; b j is the grounding admittance of node j; P jk,t,s and Q jk,t,s They represent the active power and reactive power of the line jk at the head end in the scenario s during the time period t, P ij,t,s and Q ij,t,sDefinition and P jk,t,s and Q jk,t,s similar; is the square of the voltage amplitude of node j in scenario s during time period t; P gj,t,s and Q gj,t,s represents the active output and reactive output of wind turbine g in scenario s during time period t. For the non-convex quadratic equality constraint (16d), the second-order cone relaxation method can be used to transform it into a convex constraint as shown in equation (17):

[0075]

[0076] 2) DC network power flow constraints

[0077] For offshore wind farms connected to the grid via multi-terminal VSC-HVDC, each wind farm passes through a VSC rectifier station and then forms a multi-terminal VSC-HVDC transmission system with an onshore VSC inverter station. During normal operation, its DC network has a radial structure. Therefore, similar to the branch power flow equation of the AC collector network, the DC network power flow equation can be expressed as:

[0078]

[0079]

[0080]

[0081] Where P D , and are DC network variables, and their definitions are similar to the corresponding variables of the AC network.

[0082] Equation (18c) is transformed into a convex constraint using the second-order cone relaxation method as follows:

[0083]

[0084] In order to ensure the safe operation of the wind farm collection network, it is necessary to ensure that the current flowing through the line does not exceed the maximum allowable current of the line, and also to maintain a safe voltage level, as follows:

[0085]

[0086] In the formula is the square of the maximum current amplitude allowed by the line; and Represents the square of the minimum and maximum allowed voltage amplitudes at node j.

[0087] 3) VSC converter station operation constraints

[0088] The steady-state equivalent model of the VSC converter station is as follows: Figure 4 , its internal impedance is To be equivalent, node c is a virtual node and can be included in the AC network for power flow calculation. Then the power injected into the VSC converter station and the voltage on both sides should satisfy the following formula (21):

[0089]

[0090]

[0091]

[0092]

[0093] In the formula, and They represent the active power and reactive power injected into the VSC by the AC network in scenario s during time period t respectively; and Respectively The lower and upper limits of and Respectively The lower and upper limits of is the active power output by the VSC to the DC network in scenario s during period t; and are the squares of the voltage amplitudes on the AC and DC sides of node c, respectively.

[0094] Equations (21a) to (21b) are the transmission power limits of the VSC converter station, Equation (23c) maintains the active power balance on both sides of the converter bridge, Equation (21d) is the voltage relationship on both sides of the VSC, K A is the modulation ratio, which varies in the range of [0,1], μ is the DC voltage utilization rate related to the pulse width modulation method, and when the PWM modulation method is used, it takes a constant Formula (21d) is about K A According to the nonlinear constraints of K A The range of change can simplify the original equation (21d) into the inequality constraint of equation (22):

[0095]

[0096] 4) Fan start and stop and output constraints

[0097] Fan start and stop variables and I g,t The relationship can be expressed as:

[0098]

[0099]

[0100] When the wind farm is operating normally, the output of the wind turbine must satisfy the following relationship:

[0101] P gjmin I g,t ≤P gj,t,s ≤P gjmax,t,s I g,t (25)

[0102] Q gj,t,s =P gj,t,s tanθ g (26)

[0103] Where P gjmin Indicates the minimum active output of fan g; P gjmax,t,s is the maximum active power output that can be obtained by wind turbine g in scenario s at time period t under the influence of wake effect; θ g is the power factor angle of fan g.

[0104] 5) Determination of the maximum active output that can be obtained by the units in the wind farm

[0105] The maximum active output of a single fan g can be expressed as:

[0106]

[0107] In the formula, C pg is the wind energy utilization coefficient of wind turbine g; P rated is the rated power of the fan g; ρ represents the air density; R is the rotor radius of the fan; v g is the inflow wind speed of fan g; v ci ,v rated and v co They represent the cut-in, rated and cut-out wind speeds of wind turbine g respectively. The wind power active output characteristic curve is as follows: Figure 5 shown.

[0108] To calculate the maximum active output of a single fan, you only need to know the inflow wind speed v of the fan. g However, due to the influence of the wake effect of the wind farm, the output of each wind turbine affects each other. The Jensen model is used to describe the wake effect. Assuming that wind turbines g and w are in the same straight line with the wind speed direction, it can be described as:

[0109] R wg =R+αX wg (28)

[0110]

[0111] In the formula, R wg is the wake radius generated by wind turbine w at wind turbine g along the wind speed direction;wg is the distance between wind turbines w and g in the wind speed direction; α is the wake attenuation factor, which is taken as 0.04 in offshore wind farms; C Tw is the thrust coefficient of the fan w; v0 represents the natural wind speed; v wg S represents the wake wind speed generated by wind turbine w at wind turbine g; ov,wg is the intersection area between the wake area and the wind rotor area. Assuming that wind turbine w is upstream of the wind speed and wind turbine g is downstream, the intersection area is:

[0112]

[0113] Where d represents the distance between the center of the wake area and the center of the wind rotor area. When there are G wind turbines upstream of wind turbine g, the wind speed at wind turbine g is calculated as:

[0114]

[0115] The discrete value of wind speed in each wind farm is determined in the fuzzy set. Then, the relationship between the maximum active output of each wind turbine and the wind speed under the wake effect corresponding to each discrete value can be calculated by equations (29) to (31). Assuming that wind turbine g at node j is located in wind farm m, the following equation is obtained:

[0116] P gjmax,t,s =K g,t,s v m,t,s (32)

[0117] In the formula, v m,t,s represents the wind speed value of the wind farm in time period t in the sth discrete scenario; K g,t,s It is the parameter calculated by equation (27) to equation (31).

[0118] Depend on Figure 5 It can be seen that if there is a critical wind speed v lim Satisfy v rate <v lim <v co , when the wind speed is greater than v lim When the wind turbines in the field are affected by the wake effect, they will output the rated power. Based on this feature, when establishing the fuzzy set, we can assume that v lim If it is located in the sth discrete interval, the s+1th discrete interval to the last interval can be merged into one interval.

[0119] For the convenience of representation, the DROD model is written in a compact form:

[0120]

[0121] sth(x)≤0 (33b)

[0122] g(x,y)≤0 (33c)

[0123] R(y,p)=0 (33d)

[0124] G(y)≤0 (33e)

[0125] D(y)=0 (33f)

[0126] Among them, formula (33a) represents the objective function formula (7), formula (33b) represents the start-stop constraints (23) and (24), formula (33c) represents (25), formula (33d) includes constraints (9) to (11), formula (33e) includes constraints (14), (17), (19), (20), (21a) to (21b), (22), (25), formula (33f) includes (13), (15), (16a) to (16c), (18a), (18b), (21c), (26), (32), Ψ is the fuzzy set of uncertain wind speed fluctuations, composed of formulas (1), (5) to (6). The decision variables include the start-stop state I of the wind turbine unit. g,t 、z g,t 、v g,t , Active output P g,t,s and wind speed joint probability distribution p (t,t+1),l , which is a two-stage mixed-integer nonlinear programming model, is difficult to solve directly.

[0127] Specifically, solving the optimization scheduling model includes:

[0128] (1) Column and constraint generation algorithm for solving the two-stage optimization model

[0129] For the two-stage optimization model, the C&CG algorithm solves the original problem by decomposing it into the main problem (C-MP) and the sub-problem (C-SP) alternately and iteratively, and has good convergence characteristics. The sub-problem (C-SP) is in the following form:

[0130]

[0131] The subproblem is to solve the worst distribution of wind speed p, where the start and stop variables of the wind turbine are As a known parameter in the sub-problem. Because the outer probability distribution decision p is decoupled from the inner decision y, the max-min model can be solved in two steps:

[0132]

[0133]

[0134] First solve C-SP-1 to obtain the optimal objective function value for each discrete scenario in each time period Then Pass it to C-SP-2 for solution to obtain the wind speed probability distribution that maximizes the total expected value Each time a subproblem is solved, a new set of probability distributions is obtained. And pass it to the main problem as a known parameter, adding new variables and constraints. The main problem (C-MP) is in the following form:

[0135]

[0136] Where K is the number of iterations of the CC&G algorithm, and L is the added slack variable. The main problem (C-MP) is a mixed integer second-order cone programming model, which can be solved using the GUROBI solver.

[0137] (2) Generalized Benders decomposition method for solving the main problem

[0138] Since the main problem contains a large number of discrete scenarios and wind turbine start-stop variables, and the C&CG algorithm adds new constraints and variables to the main problem in each iteration, the main problem is huge in scale. Directly solving the large-scale mixed integer SOCP model is inefficient and time-consuming. The following generalized Benders decomposition (GBD) method is used to decompose C-MP into alternating iterative solutions of MILP model and SOCP model to further improve computational efficiency. The specific decomposition steps are as follows:

[0139] (a) The main problem (GB-MP) of the GBD method is shown in Equation (38). The model only contains the wind turbine start and stop variable x, where the specific expressions of the feasible cut set and the optimal cut set will be given later.

[0140]

[0141] Here, γ is the slack variable.

[0142] (b) The sub-problem of GBD method (GB-SP) is as shown in formula (39):

[0143]

[0144] In the formula, It is the fan start and stop variable passed to GB-SP after solving the GB-MP model and is used as a fixed parameter in GB-SP. In order to obtain the feasible cut set and the optimal cut set, it is necessary to derive the dual problem of GB-SP and write it in matrix form:

[0145]

[0146] In the formula, X includes all the first-stage variables, and the superscripts indicate that they are known parameters in the model, that is, Y is a continuous variable that contains all variables except the start and stop variables. Q corresponds to the coefficient matrix of the objective function; E,b is the coefficient matrix of the equality constraint; B h ,A h ,e h ,g h ,c h ,f h is the coefficient matrix corresponding to the h-th generalized inequality.

[0147] The dual optimization problem GB-DSP of GB-SP (40) is derived as follows:

[0148]

[0149] It can be seen from formula (41) that the dual optimization model of the SOCP model is still the SOCP model. When the original problem is strictly feasible, p * =d * .

[0150] In the iterative process of GBD method, after each iteration of solving the subproblem, it is necessary to add feasible cut sets and optimal cut sets to the main problem. The feasibility test of GB-SP is to ensure the validity of the strong duality theorem, so as to ensure that the optimal objective values ​​of the original problem and its dual problem are consistent.

[0151] If GB-SP has no solution, add a feasible cut set to GB-MP as shown in formula (42):

[0152]

[0153] If GB-SP has a solution, the L-type algorithm is used to generate the optimal cut set as shown in formula (43):

[0154]

[0155] Through the GBD method, C-MP is decomposed into an alternating iterative solution of a MILP model (38) containing all 0-1 variables and a SOCP model (41) containing only continuous variables. When the solution of the subproblem satisfies the feasibility test and optimality test, the iteration converges.

[0156] (3) Algorithm steps

[0157] The proposed solution algorithm for the above model has the following algorithm flow: Figure 6 , the specific calculation steps are as follows:

[0158] 1. Initialization: Set K = 1, the lowest limit LB is -∞, and the highest limit UB is +∞. Based on the given wind speed forecast for the next day, perform deterministic optimization to obtain the initial wind turbine start and stop combination. Solve C-SP-1 and C-SP-2 in sequence to get the initial distribution

[0159] 2. Solve C-MP using GBD method:

[0160] a) Solve GB-MP to obtain the solution of the fan start-stop combination

[0161] b) Solve GB-MP to obtain It is passed to GB-SP and GB-DSP in the form of parameters for solving.

[0162] c) If GB-SP has no solution, add the feasible cut set to GB-MP and repeat step a); if the subproblem has a solution, perform an optimality test: i. If d * >γ, it means that GB-DSP has not found the optimal solution, and it is necessary to add the optimal cut set in GB-SP and repeat step 2a). ii. If d * ≤γ, indicating that the solution of GB-DSP is the optimal solution. At this time, the fan start-stop combination The optimal solution is obtained and the lowest bound is updated with the optimal objective function value of GB-MP

[0163] 3. Combine the fan start and stop Substitute into C-SP, solve C-SP-1 and C-SP-2 in turn, and get the worst distribution And use the C-SP optimal objective function value to update the highest limit

[0164] 4. Determine whether |UB-LB| / UB≤δ is satisfied. If so, the iteration ends and the optimal solution is output. Otherwise update the probability distribution of the worst wind speed in C-MP And define a new variable y K And add constraints related to the new variables. Set K = K + 1 and return to step 2 to continue iterating.

[0165] Below, this method is further explained with an application example:

[0166] Taking an offshore wind power cluster connected to the grid via multi-terminal VSC-HVDC as an example, the effectiveness of the proposed two-stage DROD model and solution algorithm is verified. The hardware environment of the test system for the example is Intel(R)Xeon(R)E3-1270 CPU@3.50GHz, 32G memory, Win1064bit operating system, and GAMS win6424.5.6 software programming.

[0167] System Parameters

[0168] The offshore wind power cluster includes three wind farms. The network topology of the cluster, the next-day dispatching plan curve issued by the power grid, the relative positions of the wind towers in each wind farm, and the relative positions of the wind turbines in each wind farm are shown in the following figure. Figures 7-10 As shown. The rated power of a single wind turbine is 5.5MW, the minimum output is 103kW, the cut-in wind speed, rated wind speed and cut-out wind speed are 3m / s, 13m / s and 25m / s respectively. The rated voltage of the AC busbar of the wind power cluster grid-connected is 220kV. The rated voltage of the AC busbars BUS-1, BUS-2 and BUS-3 on the low-voltage side of the wind farm booster station is 35kV. The DC capacitor voltage of the voltage source converter VSC is ±320kV. It is assumed that the voltage-type converters VSC1, VSC2 and VSC3 control the AC busbar voltage on the wind farm side to be stable, and VSC4 controls the DC capacitor voltage and the AC busbar voltage on the grid side to be constant. Model C-SP-2 is solved by CONOPT solver, and the other models are solved by GUROBI solver. The convergence accuracy δ of the algorithm is taken as 10 -4 .

[0169] The fuzzy set considering correlation is constructed based on the historical wind speed data of 365 days a year. The historical wind speed fluctuation range is [4.5, 21.5], and each 2m / s is divided into a sub-range. According to the wake effect and the output characteristics of the wind turbine, v is calculated. lim ∈[12.5,14.5), the combined wind speed is greater than v lim The sub-intervals are divided into 6 sub-intervals, namely [4.5, 6.5), [6.5, 8.5), [8.5, 10.5), [10.5, 12.5), [12.5, 14.5), [14.5, 21.5]. Considering that the wind speed direction at sea is mainly affected by seasonal climate, this paper takes the wind speed direction of the wind farm the next day as the dominant wind direction ENE, that is, 22.5° east by north, which remains unchanged in the decision-making process. The KL divergence threshold λ is taken as 0.1.

[0170] Computational performance analysis of the algorithm

[0171] The calculation results of the proposed C&CG algorithm for GBD solution of the main problem and the C&CG algorithm for direct solution of the main problem are compared in Table 1. In the CC&G algorithm, as the number of iterations increases, the variables and constraints of the main problem will continue to increase, which will eventually lead to a large model scale, low computational efficiency, or even failure to solve. Table 1-2 shows that the mixed integer second-order cone programming model for directly solving the main problem (C-MP) requires a lot of time due to the large number of discrete variables, especially the second iteration of solving the outer main problem (C-MP) due to the addition of variables and constraints, the time consumed increases sharply. The GBD method decomposes the main problem C-MP into alternating iterative solutions of the MILP model and the SOCP model, which effectively reduces the calculation scale of the model, thereby greatly reducing the solution time of the main problem C-MP, so that the proposed solution algorithm greatly improves the solution efficiency and can quickly and effectively solve the proposed two-stage DROD model. The total cost relative error of the calculation results of the two algorithms is less than 0.2%, indicating that the calculation results of the proposed solution algorithm are highly accurate. The outer C&CG algorithms of both algorithms converge after 2 iterations.

[0172] Table 1 Comparison of solution results of the proposed algorithm and C&CG algorithm

[0173]

[0174] Table 2 Comparison of solution results of the main problem (C-MP) in each iteration

[0175]

[0176] Analysis of DROD decision results

[0177] In order to verify that the correlation of wind speed is considered in the optimization decision, the Pearson correlation coefficient is used for testing. Table 3 shows the correlation coefficients between some wind speeds in the DROD decision results. It can be seen from the table that the corresponding wind speed correlation coefficients before and after optimization are similar, indicating that the worst joint distribution of wind speeds obtained after the decision can well preserve the correlation between wind speeds in historical data based on the fuzzy set constructed by considering the spatiotemporal correlation of wind speeds. If the fuzzy set is directly constructed without considering the correlation of wind speeds, the correlation coefficients between wind speeds corresponding to the worst distribution of wind speeds after the decision will be 0.

[0178] Table 3 Correlation coefficients between some wind speeds

[0179]

[0180] The mathematical expectation of the number of wind turbines in operation and the total active power output of each offshore wind farm in the optimization results is as follows: Fig.11 and 12As shown. It can be seen that in the time interval from the 9th period to the 22nd period, the number of wind turbines in operation in offshore wind farms 1 and 2 is equal to 91, and the expected value of the active power of offshore wind farm 2 injected into the VSC station is less than the expected value of the active power of offshore wind farm 1, as shown in Fig.12 As shown in the figure, this indicates that the probability distribution of wind speed in offshore wind farm 2 is worse, which makes the maximum active output of each wind turbine in the farm lower, so the expected value of active power in offshore wind farm 2 is lower. The number of wind turbines in operation and the expected value of active power injected into the VSC station in offshore wind farm 3 are the lowest because offshore wind farm 3 has the least number of wind turbines.

[0181] Comparison of different optimization models

[0182] The proposed DROD model is compared with the deterministic optimal dispatch model and the robust optimal dispatch (ROD) model in terms of the calculated results objective function as shown in Table 4. The wind turbine operation and cluster total active output corresponding to the calculation results of the three optimization models are compared as shown in Table 4. Figure 13-14 , where the cluster output of the DROD result is the expected value. It can be seen from Table 4 that the decision result of the proposed DROD model is more conservative than the deterministic optimization model, but more economical than the ROD model. The ROD model is optimized under the worst scenario, and the decision result obtained is very conservative. The wind turbine start-stop combination obtained by the ROD model makes all units in the wind power cluster run, which increases the wind turbine mechanical loss cost in the objective function, and the cluster power generation in the worst scenario is low, and the power generation deviation penalty cost is large, which maximizes the total cost. In fact, the worst scenario rarely occurs, resulting in unnecessary wind turbine operating costs. The proposed DROD model makes decisions under the worst distribution, and considers the wind speed probability distribution information contained in the historical data in the decision-making process, which effectively improves the shortcomings of the traditional ROD decision results that are overly conservative.

[0183] Table 4 Comparison of optimization scheduling results of different models

[0184]

[0185] Although deterministic optimization is more economical than the proposed DROD model decision, it is likely to cost more to deal with severe scenarios. The following fixed deterministic optimization model and the wind turbine start-stop combination obtained by the DROD model established in this paper, and the expected value of the worst wind speed distribution is used as the next day's wind speed forecast value, the second stage of wind turbine active output optimization is performed respectively, and the final result of each stage cost comparison is shown in Table 5. Table 5 shows that although the deterministic optimization result corresponds to a small number of wind turbines in operation and the mechanical loss cost of the wind turbines in the first stage is low, when facing a scenario worse than the predicted scenario, the power generation deviation penalty and network loss cost will be higher than the DROD result, so that the total cost is higher. Because DROD takes into account the uncertainty of wind speed, the optimized decision results have better adaptability to the uncertain fluctuations of wind speed.

[0186] Table 5 Costs of different optimization models under the expected value of the worst wind speed distribution

[0187]

[0188] In summary, the present invention establishes a two-stage distributed robust optimization scheduling model for offshore wind farm clusters connected to the grid via multi-terminal voltage source converter high voltage direct current (VSC-HVDC). The first stage of the model seeks the wind turbine start-stop combination that minimizes the total mechanical loss cost of the wind turbine under the worst wind speed distribution, and the second stage seeks the worst probability distribution in the wind speed uncertainty set and the wind turbine active output that minimizes the sum of the power generation deviation penalty cost and the collection network loss cost under this probability distribution. A data-driven fuzzy set based on the Kullback–Leibler (KL) divergence distance to describe wind speed uncertainty is proposed, and the correlation between the wind speeds of adjacent time periods in the wind farm and the correlation between the wind speeds of multiple adjacent wind farms in the cluster are reflected by adopting a joint probability distribution. A novel and efficient distributed robust optimization scheduling solution algorithm is proposed for the established optimization scheduling model. While maintaining accuracy, the algorithm has higher computational efficiency than the C&CG algorithm and meets the requirements of actual scheduling.

[0189] The above embodiments are only for illustrating the technical concept and features of the present invention, and their purpose is to enable ordinary technicians in the field to understand the content of the present invention and implement it accordingly, and they cannot be used to limit the protection scope of the present invention. Any equivalent changes or modifications made based on the essence of the content of the present invention should be included in the protection scope of the present invention.

Claims

1. A distributed robust optimization scheduling method for offshore wind power clusters considering temporal and spatial correlations. It is characterized in that include: The wind speed data of offshore wind farms is input, and the distance between the actual joint probability distribution of wind speed and the statistical reference joint probability distribution is measured by KL divergence. A data-driven fuzzy set considering the spatiotemporal correlation between wind speeds of multiple wind farms in the same cluster is constructed. A two-stage distributed robust optimization scheduling model for offshore wind power clusters connected to the grid via multi-terminal VSC-HVDC is established based on the fuzzy set. In the first stage, the wind turbine start-stop plan is optimized under the worst wind speed probability distribution to obtain the optimal wind turbine start-stop combination. In the second stage, the active output of the wind turbine is optimized. Under the given wind turbine start-stop state, the worst wind speed probability distribution that maximizes the expected value of the operating cost is obtained. Solving the optimization scheduling model and outputting the result; The fuzzy set is constructed as follows: For the wind speeds of multiple wind farms in any adjacent time period, the historical data is used to drive the generation of relevant multi-dimensional joint distributions; assuming that the vector v t represents the wind speed of M wind farms in time period t, that is, v t =(v 1,t ,...,v m,t ,...,v M,t ), where v m,t is the wind speed of wind farm m in period t; then the M-dimensional discrete joint distribution of wind speeds of M wind farms in period t p(v 1,t ,...,v m,t ,...,v M,t ) is expressed as The 2M-dimensional discrete joint distribution of wind speeds of M wind farms in time period t and time period t+1 is expressed as p(v t ,v t+1 ), then the joint probability distribution fuzzy set based on KL divergence distance is constructed as shown in formula (1): In the formula, p 0 represents the corresponding reference joint probability distribution, which can be driven by a large amount of historical data; D KL {p(v t ,v t+1 )||p 0 (v t ,v t+1 )} represents the joint probability distribution p(v t ,v t+1 ) and p 0 (v t ,v t+1 ) between the KL divergence distance; p (t,t+1),l and p 0(t,t+1),l They are respectively the true distribution p(v t ,v t+1 ) and the reference distribution p 0 (v t ,v t+1 ) is the probability value of the lth discrete scenario; N (t,t+1) is the joint distribution p(v t ,v t+1 ) is the total number of discrete value scenarios; λ is the KL divergence threshold; The sample space of the wind speed joint probability distribution among multiple wind farms in each period and the wind speed joint probability distribution of each wind farm in every two adjacent periods is statistically screened, and the sample space of the wind speed joint probability distribution among multiple wind farms in each period is statistically screened, and the sample space of the wind speed joint probability distribution of each wind farm in every two adjacent periods ... M (t),Ω m (t, t+1) represent the joint probability distribution p(v t ), the joint probability distribution of wind speed p(v m,t ,v m,t+1 ) The sample space after screening, then the joint probability distribution p(v t ,v t+1 )The elements of the sample space after screening (v 1,t ,...,v m,t ,...,v M,t ,v 1,t+1 ,...,v m,t+1 ,...,v M,t+1 ) must meet the following requirements: If the probability distribution of wind speeds in two adjacent time periods of multiple wind farms is known, p(v t ,v t+1 ), then the joint probability distribution of wind speeds of multiple adjacent wind farms in time periods t and t+1 is According to the marginal distribution law, we can obtain: In the formula, They are the joint probability distribution of wind speed of multiple wind farms in time periods t and t+1 respectively. The probability value of the sth discrete scenario; A t,l,s is the joint probability distribution of two time periods p(v t ,v t+1 ) and the single-period joint probability distribution The correlation coefficient is, when p(v t ,v t+1 ) corresponds to the lth discrete scene in When the sth discrete scenario in A t,l,s Take 1, otherwise take 0; A t+1,l,s is the joint distribution of two periods p(v t ,v t+1 ) and the single period joint distribution The correlation coefficient is the same as A t,l,s Similarly, equations (5)-(6) maintain the joint probability distribution of wind speeds of multiple adjacent wind farms in each period in each high-dimensional joint probability distribution p(v t ,v t+1 ) in consistency; The objective function of the optimization scheduling model is: In the formula, x represents the start-stop variable of the first stage, and the variables of the second stage are represented by wind speed p and wind turbine output y; E(·) represents the expected operation; T is the total number of scheduling periods; N g is the total number of fans in the cluster; a g ,b g ,c g ,d g They represent the startup cost coefficient, shutdown cost coefficient, operation cost coefficient and idle cost coefficient of fan g respectively; I g,t represents the operating state variable of fan g in period t, and "1 / 0" indicates that the fan is running / stopped; z g,t is the startup variable of fan g in time period t, "1" indicates that the fan is started, otherwise "0" is taken; u g,t is the shutdown variable of fan g in time period t, "1" indicates that the fan is shut down, otherwise "0" is taken; Considering that the probability distribution of wind speed is discrete, the expected calculation formula is obtained from the joint probability distribution of fuzzy sets: In the formula, and The joint probability distribution of multiple wind farms in time periods 1 and T are and The sth discrete probability value of is obtained by equations (5)-(6), N 1 and N T They are and The total number of scenes in the sample space, and N t yes The total number of scenes in the sample space; Q(y t,s ,v t ) including p(v t ) corresponds to the generation deviation penalty cost and network loss cost of the sth discrete scenario, C 1 and C 2 are the power generation deviation penalty cost coefficient and the network loss cost coefficient respectively; P ∑,t,s is the net active power output of the wind power cluster in scenario s during period t; P d,t It is the active power dispatching plan value sent by the power grid dispatching center to the wind power cluster; Ω E is the set of branches in the cluster; r ij is the resistance value of branch ij; is the square value of the current flowing through line ij in scenario s during time period t.

2. The offshore wind power cluster distributed robust optimization scheduling method considering time and space correlation as claimed in claim 1, It is characterized in that The solving of the optimization scheduling model comprises: The C&CG algorithm is used to divide the two-stage robust optimization scheduling model into a main problem and sub-problems for alternating iterative solution, and the generalized Benders decomposition method is used to decompose the main problem into a mixed integer linear programming model containing all 0-1 variables and a continuous second-order cone programming model for alternating iterative solution.

3. The offshore wind power cluster distributed robust optimization scheduling method considering time and space correlation as claimed in claim 1, It is characterized in that Formula (13) to (15) are used to replace formula (12) containing absolute value calculation.

4. The offshore wind power cluster distributed robust optimization scheduling method considering time and space correlation as claimed in claim 1 or 3, It is characterized in that The constraints of the optimization scheduling model include AC collection network operation constraints, DC network flow constraints, VSC converter station operation constraints, wind turbine start and stop and output constraints, and determination of the maximum active output that can be obtained by the units in the wind farm.

5. The offshore wind power cluster distributed robust optimization scheduling method considering time and space correlation as claimed in claim 4, It is characterized in that The AC collection network operation constraints include: The AC collection network of an offshore wind farm is usually a radial network, which can be described by a branch flow model as follows: In the formula, δ(j) / π(j) represents the set of nodes whose parent node / child node is j; x ij is the reactance value of line ij; b j is the grounding admittance of node j; P jk,t,s and Q jk,t,s They represent the active and reactive power of the line jk at the head end in the scenario s during the time period t, P ij,t,s and Q ij,t,s Definition and P jk,t,s and Q jk,t,s similar; is the square of the voltage amplitude of node j in scenario s during time period t; P gj,t,s and Q gj,t,s represents the active output and reactive output of wind turbine g in scenario s during time period t; for the non-convex quadratic equality constraint (16d), the second-order cone relaxation method can be used to transform it into a convex constraint as shown in formula (17): The DC network power flow constraints include: Where P D , and are DC network variables, and their definitions are similar to the corresponding variables of the AC network; Equation (18c) is transformed into a convex constraint using the second-order cone relaxation method as follows: In order to ensure the safe operation of the wind farm collection network, it is necessary to ensure that the current flowing through the line does not exceed the maximum allowable current of the line, and also to maintain a safe voltage level, as follows: In the formula is the square of the maximum current amplitude allowed by the line; and Represents the square of the minimum and maximum allowed voltage amplitudes at node j.

6. The offshore wind power cluster distributed robust optimization scheduling method considering time and space correlation as claimed in claim 4, It is characterized in that The VSC converter station operation constraints include: A steady-state equivalent model of the VSC converter station is established, and its internal impedance is expressed as To be equivalent, node c is a virtual node and is included in the AC network for power flow calculation. Then the power injected into the VSC converter station and the voltage on both sides should satisfy the following formula (21): In the formula, and They represent the active power and reactive power injected into the VSC by the AC network in scenario s during time period t respectively; and Respectively The lower and upper limits of and Respectively The lower and upper limits of is the active power output by the VSC to the DC network in scenario s during period t; and are the squares of the voltage amplitudes on the AC and DC sides of node c, respectively; Equations (21a) to (21b) are the transmission power limits of the VSC converter station, Equation (23c) maintains the active power balance on both sides of the converter bridge, Equation (21d) is the voltage relationship on both sides of the VSC, K A is the modulation ratio, ranging from [0,1], μ is the DC voltage utilization rate related to the pulse width modulation method; Formula (21d) is about K A According to the nonlinear constraints of K A The range of change simplifies the original equation (21d) into the inequality constraint of equation (22):

7. The offshore wind power cluster distributed robust optimization scheduling method considering time and space correlation as claimed in claim 4, It is characterized in that The wind turbine start / stop and output constraints include: Fan start and stop variables and I g,t The relationship is expressed as: When the wind farm is operating normally, the output of the wind turbine must satisfy the following relationship: P gjmin ·I g,t ≤P gj,t,s ≤P gjmax,t,s ·I g,t (25) Q gj,t,s =P gj,t,s ·tanθ g (26) Where P gjmin Indicates the minimum active output of fan g; P gjmax,t,s is the maximum active power output that can be obtained by wind turbine g in scenario s at time period t under the influence of wake effect; θ g is the power factor angle of fan g.

8. The offshore wind power cluster distributed robust optimization scheduling method considering time and space correlation as claimed in claim 7, It is characterized in that The determination of the maximum active output that can be obtained by the units in the wind farm includes: The maximum active output of a single fan g can be expressed as: In the formula, C pg is the wind energy utilization coefficient of wind turbine g; P rated is the rated power of the fan g; ρ represents the air density; R is the rotor radius of the fan; v g is the inflow wind speed of fan g; v ci ,v rated and v co They represent the cut-in, rated and cut-out wind speeds of fan g respectively.