Active power control method and system for wind power plant

Abstract

The invention relates to the technical field of wind power plant active power control, and discloses an active power control method and system for a wind power plant, and the method comprises the steps: obtaining the historical data of a fan, and constructing a training set according to a wake flow influence result of a fan state; based on an AdaBoost algorithm, constructing a strong classifier in combination with the training set; the method comprises the following steps: determining fans influenced by wake flow by using a strong classifier, determining upstream fans generating wake flow by using a wake flow center line deflection angle, and defining all wind turbine generators which are not influenced by wake flow and generate wake flow which does not influence downstream fans as a controllable fan group; and solving the wind turbine generator groups with the minimum number of execution actions in the wind power plant by utilizing a dynamic programming algorithm, and executing active power control operation. According to the method, the space-time coupling influence caused by the wake effect in the active power control process of the wind power plant can be eliminated, the environmental disturbance generated by the execution of the power control action of the fan to the wind power plant is reduced to the minimum, and the real-time accurate control of the active power of the wind power plant can be realized.

Description

Technical Field

[0001] The present invention relates to the technical field of power control, and in particular to an active power control method and system for a wind farm. Background Art

[0002] At present, the active power control method of wind farms is to allocate each unit to perform active power control according to the active power control amount issued by the dispatcher and the actual controllable capacity of each unit in the wind farm. However, due to the existence of the wake effect, the wake generated by the wind turbine at the upwind position will affect the output power of the wind turbine at the downwind position, and there is a temporal and spatial coupling between the two. During the execution of active power control, the wake changes caused by the power control action of the upstream wind turbine cause the output active power of the downstream wind turbine to fluctuate, reducing the overall active power control accuracy of the wind farm.

[0003] The existing wake calculation models are mainly divided into analytical wake models and computational wake models. The former is a mathematical expression of the velocity distribution in the wake area of ​​a semi-empirical wind turbine based on idealized assumptions or experimental data fitting; the latter uses a complete computational fluid dynamics analysis method to solve the N·S equation to more accurately describe the flow of the wind turbine wake field. The defects of the existing technology are as follows:

[0004] 1. The mathematical expression of the velocity distribution in the semi-empirical wind turbine wake area obtained by analytical wake model based on idealized assumptions or experimental data fitting often cannot meet the accuracy requirements of on-site engineering. For active power control of wind farms, the establishment of a dynamic wake model for the entire wind farm requires a large amount of calculation and has a slow response speed.

[0005] 2. The computational wake model uses a complete computational fluid dynamics analysis method to solve the N·S equation to more accurately describe the flow of the wind turbine wake field. The calculation speed is also slow and cannot meet the real-time requirements on site.

[0006] 3. Due to operation and maintenance and site restrictions, it is difficult to deploy sensors in wind farms and wind turbines that have been put into operation to obtain the parameters of the above two models to analyze the wake state.

[0007] Currently, no effective solution has been proposed for the problems in the related technologies. Summary of the invention

[0008] In view of the problems in the related art, the present invention proposes an active power control method and system for a wind farm to overcome the above technical problems existing in the existing related art.

[0009] To this end, the specific technical solution adopted by the present invention is as follows:

[0010] According to one aspect of the present invention, there is provided a method for controlling active power of a wind farm, comprising the following steps:

[0011] S1. Obtain historical data of the wind turbine and construct a training set based on the results of the wind turbine status being affected by the wake; construct a strong classifier based on the AdaBoost algorithm and the training set;

[0012] S2. Use a strong classifier to determine the wind turbines affected by the wake, use the deflection angle of the wake centerline to determine the upstream wind turbines that generate the wake, and define all wind turbines that are not affected by the wake and whose wake does not affect the downstream wind turbines as controllable wind turbine groups;

[0013] S3. Based on the controllable capacity of each wind turbine in the controllable wind turbine group, a dynamic programming algorithm is used to solve the wind turbine group with the least number of execution actions in the wind farm, and the wind turbines in the group are used as the final execution wind turbines to perform active power control operations based on the controllable capacity of each wind turbine.

[0014] Furthermore, the method of obtaining historical data of the wind turbine and constructing a training set according to the result of the wind turbine state being affected by the wake; and constructing a strong classifier based on the AdaBoost algorithm and the training set comprises the following steps:

[0015] S11, obtaining historical data of the wind turbine, and determining whether the state of the wind turbine corresponding to the historical data is affected by the wake according to the change of the deviation between the ambient wind direction and the local wind direction with the wind direction angle, wherein the historical data includes historical operation data, environmental historical data and meteorological historical data;

[0016] S12, combining historical data and the results of whether the data is affected by wake turbulence to obtain a training set; initializing the weight distribution of each sample in the training set, and adjusting the weight of each sample according to whether the sample classification is correct;

[0017] S13, based on the weight distribution, establish a weak classifier according to the attribute characteristics of the historical operation data, the environmental historical data and the meteorological historical data, and select the value that minimizes the classification error of the training set as the threshold of the weak classifier;

[0018] S14. Based on the weight coefficient formula, the weight of the weak classifier is calculated, and the weight distribution of the training set is updated according to the weight of the weak classifier to obtain a new weight distribution of the sample;

[0019] S15. Calculate the judgment value of the overall classification according to the weighted average of each weak classifier, and construct a final strong classifier based on the classification result of the judgment value.

[0020] Furthermore, the combination of historical data and the result of whether it is affected by wake turbulence to obtain a training set; initializing the weight distribution of each sample in the training set, and adjusting the weight of each sample according to whether the sample classification is correct or not includes:

[0021] Combining historical data and the results of whether it is affected by wake turbulence, we get the training set M = {(x1, y1), (x2, y2), ..., (x i ,y i ),…,(x m ,y m )};

[0022] Initialize the weight distribution D1 of each sample in the training set = (ω 11 ,,ω 12 ,…,ω 1m ),ω 1i =1 / m;

[0023] Determine whether each sample in the training set is accurately classified. If so, reduce the weight of the sample when constructing the next training set. If not, increase the weight of the sample when constructing the next training set.

[0024] In the formula, x i represents the i-th historical data, y i represents the wake influence state corresponding to the i-th historical data, m represents the number of samples, D1 represents the weight of the training set at the first iteration, ω 1i Represents the weight of the i-th historical data in the first iteration.

[0025] Furthermore, the expression of the weak classifier is:

[0026] h t (x) = ξ(M, D t )

[0027] The expression of classification error is:

[0028]

[0029] In the formula, h t (x) represents the tth weak classifier, ξ(M, D t ) represents the classifier function, D t represents the weight of the training set at the tth iteration, e t represents the tth weak classifier h t (x) will have weight D t The classification error after the classification of the data set M, ω ti represents the weight of the i-th sample in the t-th iteration, h t (x i ) means that the i-th historical data is classified according to the classifier h t The classification result is obtained by dividing (x), and I represents the integer weight function.

[0030] Furthermore, the weight coefficient formula is:

[0031]

[0032] The expression for updating the weight distribution of the training set is:

[0033] D t+1 =(ω (t+1)1 ,,ω (t+1)2 ,…,ω (t+1)m )

[0034]

[0035] In the formula, α t Represents the classifier h t (x) Weight coefficient in the final classifier, D t+1 represents the weight of the training set M at the t+1th iteration, ω (t+1)i represents the weight of the i-th historical data in the t+1th iteration, Z t represents the normalization factor.

[0036] Furthermore, the expression of the judgment value is:

[0037]

[0038] The final strong classifier expression is:

[0039]

[0040] In the formula, f(x) represents the judgment value of each weak classifier combined with the weight coefficient to classify the historical data, t represents the number of iterations, T represents the total number of iterations, α t Represents the classifier h t (x) is the weight coefficient in the final classifier, h t (x) represents the tth weak classifier, h(x) represents the final strong classifier, and sign(f(x)) represents the judgment value classification function.

[0041] Further, the method of using a strong classifier to determine the wind turbines affected by the wake, using the wake centerline deflection angle to determine the upstream wind turbines generating the wake, and defining all wind turbines that are not affected by the wake and whose wake does not affect the downstream wind turbines as a controllable wind turbine group includes the following steps:

[0042] S21, using a strong classifier to determine the wind turbine affected by the wake, and to determine all upstream wind turbines within a preset range upwind of the wind turbine;

[0043] S22, calculating the wake centerline deflection angles of all upstream wind turbines within a preset range based on the expression of the wake centerline deflection angle;

[0044] S23, determining whether the wake centerline deflection angle of the upstream wind turbine is equal to the angle difference between the local wind direction of the wind turbine and the ambient wind direction; if so, determining that the upstream wind turbine is an upstream wind turbine that generates wake influence; if not, determining that the upstream wind turbine is not an upstream wind turbine that generates wake influence;

[0045] S24. All wind turbines that are not affected by wake vortices and whose wake vortices do not affect downstream wind turbines are defined as controllable wind turbine groups.

[0046] Furthermore, the expression of the wake centerline deflection angle is:

[0047]

[0048]

[0049] Where δ(x) represents the deflection angle of the centerline of the upstream wind turbine wake, c p represents the yaw wind power coefficient of the upstream wind turbine, x represents the distance between the upstream wind turbine and the current wind turbine, k d represents the wake diffusion coefficient, D represents the fan impeller diameter, θ represents the yaw angle of the upstream fan, C t Represents the thrust coefficient of the upstream wind turbine rotor.

[0050] Furthermore, based on the controllable capacity of each wind turbine in the controllable wind turbine group, a dynamic programming algorithm is used to solve the wind turbine group with the least number of execution actions in the wind farm, and the wind turbines in the wind turbine group are used as the final execution wind turbines. The active power control operation based on the controllable capacity of each wind turbine includes the following steps:

[0051] S31. Obtain the controllable capacity p of each fan in the controllable fan group i , and establish a set of controllable capacities in the controllable wind turbine group, determine the target active power control amount P0; construct two arrays for storing the minimum number of wind turbines d[P] and wind turbine numbers required to achieve each active power control amount;

[0052] Among them, the expression of the set of controllable capacities is:

[0053] P 可控 ={p1, p2, ..., p i , ..., p n}

[0054] Where P 可控 represents the set of controllable capacities in the controllable wind turbine group, n represents the number of wind turbines in the controllable wind turbine group, p i represents the controllable capacity of the i-th fan in the controllable fan group;

[0055] S32, initializing the minimum number of wind turbines required when the target active power control amount is zero d[0]=0; calculating the minimum number of wind turbines required for each active power control amount based on the state transfer equation;

[0056] Among them, the expression of the state transfer equation is:

[0057]

[0058] Where d[P0] represents the minimum number of wind turbines required for the target active power control quantity P0, d[P0-p i ] indicates that the target active power control amount is P0-p i The minimum number of wind turbines required;

[0059] S33, gradually increase the active power control amount from zero, calculate the value of the minimum number of wind turbines d[P] required for the active power control amount of each sub-problem, and record the optimal solution of each state and the number of wind turbines that reach the state;

[0060] S34, iteratively calculate until the target active power control amount P0 is reached, and output the minimum number of wind turbine groups and their numbers required to achieve the target active power control amount P0.

[0061] According to another aspect of the present invention, there is provided an active power control system for a wind farm, the active power control system for the wind farm comprising a strong classifier building module, a controllable wind turbine group determination module and a dynamic programming solution module;

[0062] The strong classifier construction module is used to obtain historical data of the wind turbine and construct a training set according to the result of the wind turbine state being affected by the wake; based on the AdaBoost algorithm, a strong classifier is constructed in combination with the training set;

[0063] The controllable wind turbine group determination module is used to determine the wind turbines affected by the wake using a strong classifier, determine the upstream wind turbines generating the wake using the wake centerline deflection angle, and define all wind turbines that are not affected by the wake and whose wake does not affect the downstream wind turbines as the controllable wind turbine group;

[0064] The dynamic programming solution module is used to solve the wind turbine group with the least number of execution actions in the wind farm based on the controllable capacity of each wind turbine in the controllable wind turbine group using a dynamic programming algorithm, and use the wind turbines in the group as the final execution wind turbines to perform active power control operations based on the controllable capacity of each wind turbine.

[0065] The beneficial effects of the present invention are:

[0066] 1) The present invention uses existing wind turbine operation data, environmental data, and meteorological data in a non-invasive manner to identify and eliminate the spatiotemporal coupling effects caused by wake effects during active power control. First, a strong classifier is constructed using the AdaBoost algorithm to identify whether the wind turbine is affected by the wake. If it is affected by the wake, the upstream wind turbine that generates the wake is determined based on the coordinate position and the wake centerline deflection formula. Secondly, by identifying whether the wind turbine is affected by the wake and whether it affects the downstream wind turbine, a controllable wind farm group is constructed for the wind turbines that are not affected by the spatiotemporal coupling of the wake effect, and the actual controllable capacity of each wind turbine in the group is determined. Finally, a bottom-up recursive dynamic programming algorithm is used to calculate the minimum number of executing wind turbines required to achieve the target active power control amount, and active power control operations are performed based on the controllable capacity of each wind turbine in the group. This method can eliminate the spatiotemporal coupling effects caused by the wake effect during the active power control process of the wind farm, and minimize the environmental disturbance caused by the power control action of the executing wind turbine to the wind farm. It has a clear physical meaning and improves the control accuracy of the active power of the wind farm.

[0067] 2) The present invention uses the AdaBoost algorithm to construct a strong classifier to identify whether the wind turbine is affected by the wake. Compared with the traditional wake model calculation method, this method does not require the addition of additional sensor arrays and other equipment, is not restricted by the actual operation and maintenance site, and uses the existing operation data, environmental data and meteorological data to identify and process the wake effect, providing a new solution for the upgrade and transformation of the old system.

[0068] 3) The method of identifying the wind turbine wake effect by using a strong classifier in the present invention has a faster operation speed and higher classification accuracy than the traditional analytical wake model and the computational wake model. The number of classifiers can be controlled according to the actual classification accuracy requirements on site, the performance of the on-site server and other actual conditions, the amount of calculation can be flexibly adjusted, and the relationship between the operation speed and the classification accuracy can be balanced. Compared with the traditional wind farm active power control strategy, the present invention requires the least number of wind turbines to execute power control, reduces the amount of calculation at the control layer, and can better meet the real-time requirements of active power control in wind farms.

[0069] 4) The present invention adopts a bottom-up recursive dynamic programming algorithm. When solving the minimum number of wind turbines, each sub-problem is solved only once, and then the solution is saved in a table. When solving this sub-problem again, it only takes a constant time to check the result. Usually, the number of sub-problems increases polynomially with the increase of the problem scale. The algorithm only takes a constant time, thereby ensuring the real-time requirements of active power control of wind farms. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0071] Figure 1 is a flow chart of an active power control method of a wind farm according to an embodiment of the present invention;

[0072] Figure 2 It is a schematic diagram of the principle of a recursive dynamic programming algorithm in a method for controlling active power of a wind farm according to an embodiment of the present invention. DETAILED DESCRIPTION

[0073] To further illustrate each embodiment, the present invention provides drawings, which are part of the disclosure of the present invention and are mainly used to illustrate the embodiments and can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. With reference to these contents, ordinary technicians in the field should be able to understand other possible implementations and advantages of the present invention. The components in the figures are not drawn to scale, and similar component symbols are generally used to represent similar components.

[0074] According to an embodiment of the present invention, a method and system for controlling active power of a wind farm are provided.

[0075] The present invention will now be further described with reference to the accompanying drawings and specific embodiments. Figure 1-Figure 2 As shown, according to an embodiment of the present invention, a method for controlling active power of a wind farm is provided, comprising the following steps:

[0076] S 1. Obtain historical data of the wind turbine and construct a training set based on the results of the wind turbine status being affected by the wake; build a strong classifier based on the AdaBoost algorithm and the training set;

[0077] In the present embodiment, Adaboost is an ensemble learning algorithm, which adds a new weak classifier in each round of iteration until a predetermined accuracy is reached. Its core idea is to train different weak classifiers for the same training set, and then combine these weak classifiers to form a stronger final classifier. The algorithm determines the weight of each sample based on whether the classification of each sample in the training set is correct and the accuracy of the last overall classification. The training set with modified weights is sent to the next classifier for training. Finally, the classifiers obtained from each training are merged as the final decision classifier.

[0078] Specifically, the method of acquiring historical data of the wind turbine and constructing a training set according to the result of the wind turbine state being affected by the wake; and constructing a strong classifier based on the Adaboost algorithm and the training set comprises the following steps:

[0079] S11. Obtain historical data of the wind turbine, including historical operation data, environmental historical data, and meteorological historical data. According to the wake effect, the deviation between the ambient wind direction and the local wind direction will change with the wind direction angle, showing positive and negative alternating waveforms, etc., and determine whether the state of the wind turbine corresponding to the historical data is under the influence of the wake;

[0080] S12, combining historical data and the results of whether the data is affected by wake turbulence to obtain a training set; initializing the weight distribution of each sample in the training set, and adjusting the weight of each sample according to whether the sample classification is correct;

[0081] Specifically, the combination of historical data and the result of whether it is affected by wake turbulence to obtain a training set; initializing the weight distribution of each sample in the training set, and adjusting the weight of each sample according to whether the sample classification is correct or not includes:

[0082] Combining historical data and the results of whether it is affected by wake turbulence, we get the training set:

[0083] M={(x1, y1), (x2, y2),..., (x i ,y i ),...,(x m ,y m )}

[0084] In the formula, x i represents the i-th historical data, y i Indicates the wake influence state corresponding to the i-th historical data. If the wind turbine is affected by the wake y i =1, if the fan is not affected by the wake i =-1, m represents the number of samples;

[0085] Initialize the weight distribution D1 of each sample in the training set. There are m samples in the training set M. At this time, the weight of each training sample is assigned the same value 1 / m:

[0086] D1=(ω 11 ,,ω 12 ,…,ω 1m ),ω 1i =1 / m

[0087] Where D1 represents the weight of the training set M at the first iteration, ω 1i Represents the weight of the i-th historical data in the first iteration;

[0088] If a sample has been accurately classified, its weight is reduced when constructing the next training set; conversely, if a sample is misclassified, its weight is increased;

[0089] S13, let the number of iterations be t, t = 1, 2, ..., T. Based on the weight distribution D t Train a weak classifier h from the dataset M t (x) (establish a weak classifier based on the attribute characteristics of the operating data, environmental data and meteorological data, and select the value that minimizes the classification error of the training set as the threshold of the classifier);

[0090] Wherein, the expression of the weak classifier is:

[0091] h t (x) = ξ(M, D t )

[0092] The expression of classification error is:

[0093]

[0094] In the formula, h t (x) represents the tth weak classifier; ξ(M, D t ) represents the classifier function, which represents the classifier with weight D t Each historical data in the dataset M is divided into two categories according to the threshold of a certain attribute feature, namely, affected by wake and not affected by wake. The threshold based on which the classification is based is the threshold with the smallest classification error among all thresholds; D t represents the weight of the training set at the tth iteration; e t represents the tth weak classifier h t (x) will have weight D t The classification error after the classification of the data set M; ω ti represents the weight of the i-th sample in the t-th iteration; h t (x i ) means that the i-th historical data is classified according to the classifier h t The classification result obtained by the division method of (x); I represents the integer weight function, h t (x i ) is 0 if the classification is correct, h t (x i ) is 1 if the classification is wrong.

[0095] From the above formula, we can see that h t (x) Classification error e on the training data set t is h t (x) The sum of the weights of misclassified samples.

[0096] S14, based on the weight coefficient formula, calculate the weight of the weak classifier, and update the weight distribution of the training set according to the weight of the weak classifier to obtain a new weight distribution of the sample for the next round of iteration;

[0097] Wherein, the weight coefficient formula is:

[0098]

[0099] The expression for updating the weight distribution of the training set is:

[0100] D t+1 =(ω (t+1)1 ,,ω (t+1)2 ,…,ω (t+1)m )

[0101]

[0102] In the formula, α t Represents the classifier h t (x) is the weight coefficient in the final classifier. The ln function is a logarithmic function with the mathematical constant e as the base. t+1 represents the weight of the training set M at the t+1th iteration, exp is an exponential function with the mathematical constant e as the base, ω (t+1)i represents the weight of the i-th historical data in the t+1th iteration, Z t represents the normalization factor.

[0103] From the above formula, we can conclude that D t+1 is a probability distribution. Update the weight distribution of the training set so that it is t (x) The weight of misclassified samples increases, while the weight of correctly classified samples decreases. In this way, the AdaBoost algorithm can focus on samples that are more difficult to classify.

[0104] S15. Calculate the judgment value of the overall classification according to the weighted average of each weak classifier, and construct a final strong classifier based on the classification result of the judgment value (ie, classify the judgment value into -1 and 1 to obtain the final strong classifier).

[0105] The expression of the judgment value is:

[0106]

[0107] The final strong classifier expression is:

[0108]

[0109] In the formula, f(x) represents the judgment value of each weak classifier combined with the weight coefficient to classify the historical data, t represents the number of iterations, T represents the total number of iterations, α t Represents the classifier h t (x) is the weight coefficient in the final classifier, h i (x) represents the tth weak classifier, h(x) represents the final strong classifier, sign(f(x)) represents the judgment value classification function, when x≥0, sign(f(x))=1; when x<0, sign(f(x))=-1;

[0110] The total number of iterations T is determined according to on-site requirements until the final strong classifier h(x) reaches a predetermined classification accuracy.

[0111] S2. Use a strong classifier to determine the wind turbines affected by the wake, use the deflection angle of the wake centerline to determine the upstream wind turbines that generate the wake, and define all wind turbines that are not affected by the wake and whose wake does not affect the downstream wind turbines as controllable wind turbine groups;

[0112] Specifically, whether the wind turbine is affected by the wake is determined according to the strong classifier. If affected by the wake, the upstream wind turbine generating the wake is determined according to the wind direction and coordinate information.

[0113] When the upstream wind turbine yaws and causes the wake to deflect, the influence range of the wind turbine wake on the spanwise side (the side facing after the wind rotor yaws) increases, but its main influence range is still limited to the 2D range of the wind rotor spanwise (D is the wind rotor diameter). The actual wind turbine spacing in the wind farm is 4D-10D. In the actual wind farm, due to the influence of turbulent inflow, even in the non-yaw state, the wake velocity loss is relatively small at about 8D downstream. Therefore, when considering the influence of the wake, it is only necessary to find the upstream wind turbine closest to the upwind direction.

[0114] In this embodiment, assuming that a certain wind turbine W (ie, the current wind turbine) is affected by the wake, firstly, all upstream wind turbines within a range of 8D in the upwind direction are determined, and then the wake centerline deflection angle of the upstream wind turbine is calculated.

[0115] The method of using a strong classifier to determine the wind turbines affected by the wake, using the deflection angle of the wake centerline to determine the wind turbines generating the wake upstream, and defining all wind turbines that are not affected by the wake and whose wake does not affect the downstream wind turbines as a controllable wind turbine group includes the following steps:

[0116] S21, using a strong classifier to determine the wind turbine affected by the wake, and to determine all upstream wind turbines within a preset range upwind of the wind turbine;

[0117] S22, calculating the wake centerline deflection angles of all upstream wind turbines within a preset range based on the expression of the wake centerline deflection angle;

[0118] Specifically, the expression of the wake centerline deflection angle is:

[0119]

[0120] The wake diffusion coefficients of the near wake area, mixed wake area and far wake area are as follows:

[0121]

[0122] Where δ(x) represents the deflection angle of the centerline of the upstream wind turbine wake, c p represents the upstream wind turbine yaw power coefficient, k d represents the wake diffusion coefficient, x is the distance between the upstream fan and the fan W, D is the fan impeller diameter, θ is the yaw angle of the upstream fan, C t Represents the thrust coefficient of the upstream wind turbine rotor.

[0123] S23, determining whether the wake centerline deflection angle of the upstream wind turbine is equal to the angle difference between the local wind direction of the wind turbine and the ambient wind direction; if so, determining that the upstream wind turbine is an upstream wind turbine that generates wake influence; if not, determining that the upstream wind turbine is not an upstream wind turbine that generates wake influence;

[0124] S24. All wind turbines that are not affected by wake vortices and whose wake vortices do not affect downstream wind turbines are defined as controllable wind turbine groups.

[0125] Specifically, due to the existence of the wake effect, the wake generated by the wind turbine at the upwind position will affect the output power of the wind turbine at the downwind position, and there is a spatiotemporal coupling between the two. During the execution of active power control, the wake changes caused by the power control action of the upstream wind turbine cause the active power output of the downstream wind turbine to fluctuate. When the downstream wind turbine affected by the wake performs the control action, the local wind speed, wind direction, air pressure and other environmental factors of the wind turbine will produce nonlinear changes due to the wake turbulence, affecting the control accuracy. The above two factors jointly reduce the overall active power control accuracy of the wind farm. Therefore, in this embodiment, the wind turbines that are not affected by the wake and whose wake does not affect the downstream wind turbines will form a controllable wind turbine group.

[0126] S3. Based on the controllable capacity of each wind turbine in the controllable wind turbine group, a dynamic programming algorithm is used to solve the wind turbine group with the least number of execution actions in the wind farm, and the wind turbines in the group are used as the final execution wind turbines to perform active power control operations based on the controllable capacity of each wind turbine.

[0127] In this embodiment, in order to minimize the impact of power control actions on the environmental disturbance of the wind farm and improve the control accuracy of the active power of the wind farm, a bottom-up recursive dynamic programming idea is used to solve the minimum number of executing wind turbines. The dynamic programming algorithm solves each subproblem only once, and then saves its solution in a table. When solving this subproblem again, it only takes a constant time to check the result. Usually, the number of subproblems grows polynomially with the increase of the problem scale. The dynamic programming algorithm only takes a constant time, thereby achieving a higher problem-solving efficiency.

[0128] The method of performing active power control based on the controllable capacity of each wind turbine in the controllable wind turbine group, using a dynamic programming algorithm to solve the wind turbine group with the least number of wind turbines in the wind farm, and taking the wind turbines in the wind turbine group as the final execution wind turbines, comprises the following steps:

[0129] If the controllable wind turbine group has n wind turbines in total, obtain the controllable capacity of each wind turbine in the controllable wind turbine group:

[0130] P 可控 ={p1, p2, ..., p i , ..., p n}

[0131] Where P 可控 represents the set of controllable capacities in the controllable wind turbine group, n represents the number of wind turbines in the controllable wind turbine group, p i It represents the controllable capacity of the i-th wind turbine in the controllable wind turbine group.

[0132] Assume that the target active power control quantity assigned by the dispatcher is P0, and the minimum number of wind turbines required to form the active power control quantity P0 is d[P0]. The bottom-up recursive dynamic programming algorithm constructs two arrays to store the minimum number of wind turbines and wind turbine numbers required to achieve each active power control quantity. Each time a subproblem needs to be calculated, first determine whether the result of the subproblem exists in the array. If so, use the result in the array directly. If not, calculate and save the result in the array. Start from the active power control quantity 0 and gradually build a solution. Using the optimal substructure property of the problem, recursively construct the optimal solution of the entire problem from the optimal solution of the subproblem in a bottom-up manner until the active power control quantity P0 is reached;

[0133] Establish the state transfer equation:

[0134]

[0135] Where d[P0] represents the minimum number of wind turbines required for the target active power control quantity P0, d[P0-p i ] indicates that the target active power control amount is P0-pi The minimum number of wind turbines required;

[0136] initialization:

[0137] d[0]=0

[0138] If the target active power control amount is P0, then the active power control amount P0 can be obtained by the active power control amount P0-p i Plus an active power control quantity p i If we know d[P0-p i ], we can add a machine with active power control quantity p i To obtain d[P0], we need to consider all active power control quantities less than P0, and then add the active power control quantity of a certain wind turbine. According to the solutions to these sub-problems, the optimal solution can be obtained by selecting the solution with the least number of wind turbines to execute, and output the wind turbine group and its number with the least number of actions required to achieve the target active power control quantity P0.

[0139] The bottom-up recursive dynamic programming algorithm solves the original problem by building solutions to sub-problems and ensures that each step is optimized. This method can find the minimum number of wind turbine groups that execute actions corresponding to each active power control quantity.

[0140] According to another embodiment of the present invention, there is provided an active power control system of a wind farm, the active power control system of the wind farm comprising a strong classifier building module, a controllable wind turbine group determination module and a dynamic programming solution module;

[0141] The strong classifier construction module is used to obtain historical data of the wind turbine and construct a training set according to the result of the wind turbine state being affected by the wake; based on the AdaBoost algorithm, a strong classifier is constructed in combination with the training set;

[0142] The controllable wind turbine group determination module is used to determine the wind turbines affected by the wake using a strong classifier, determine the upstream wind turbines generating the wake using the wake centerline deflection angle, and define all wind turbines that are not affected by the wake and whose wake does not affect the downstream wind turbines as the controllable wind turbine group;

[0143] The dynamic programming solution module is used to solve the wind turbine group with the least number of execution actions in the wind farm based on the controllable capacity of each wind turbine in the controllable wind turbine group using a dynamic programming algorithm, and use the wind turbines in the group as the final execution wind turbines to perform active power control operations based on the controllable capacity of each wind turbine.

[0144] In summary, with the help of the above technical solution of the present invention, the present invention uses existing wind turbine operation data, environmental data, and meteorological data in a non-invasive manner to realize the identification and elimination of the spatiotemporal coupling effects caused by the wake effect in the active power control process. First, a strong classifier is constructed by the AdaBoost algorithm to identify whether the wind turbine is affected by the wake. If it is affected by the wake, the upstream wind turbine that generates the wake is determined according to the coordinate position and the deflection formula of the wake centerline. Secondly, by identifying whether the wind turbine is affected by the wake and whether it affects the downstream wind turbine, the wind turbines that are not affected by the spatiotemporal coupling of the wake effect are constructed into a controllable group of wind farms, and the actual controllable capacity of each wind turbine in the group is determined. Finally, a bottom-up recursive dynamic programming algorithm is used to calculate the minimum number of execution wind turbines required to achieve the target active power control amount, and the active power control operation is performed based on the controllable capacity of each wind turbine in the group. This method can eliminate the influence of time and space coupling caused by the wake effect during the active power control process of the wind farm, and minimize the environmental disturbance caused by the power control action of the wind turbine on the wind farm. It has clear physical meaning and improves the control accuracy of the active power of the wind farm.

[0145] In addition, the present invention uses the AdaBoost algorithm to build a strong classifier to identify whether the wind turbine is affected by the wake. Compared with the traditional wake model calculation method, this method does not require the addition of additional sensor arrays and other equipment, is not restricted by the actual operation and maintenance site, and uses existing operating data, environmental data and meteorological data to identify and process the wake effect, providing a new solution for the upgrade and transformation of old systems.

[0146] In addition, the method of identifying the wake effect of wind turbines by using a strong classifier in the present invention has a faster operation speed and higher classification accuracy than traditional analytical wake models and computational wake models. The number of classifiers can be controlled according to actual classification accuracy requirements on site, on-site server performance and other actual conditions, and the amount of calculation can be flexibly adjusted to balance the relationship between operation speed and classification accuracy. Compared with traditional wind farm active power control strategies, the present invention requires the least number of wind turbines to execute power control, reduces the amount of calculation at the control layer, and can better meet the real-time requirements of active power control in wind farms.

[0147] In addition, the present invention adopts a bottom-up recursive dynamic programming algorithm. When solving the minimum number of wind turbines, each sub-problem is solved only once, and then the solution is saved in a table. When solving this sub-problem again, it only takes a constant time to check the result. Usually, the number of sub-problems increases polynomially with the increase of the problem scale. The algorithm only takes a constant time, thereby ensuring the real-time requirements of active power control of wind farms.

[0148] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for controlling active power of a wind farm, characterized in that: The active power control method of the wind farm comprises the following steps: S 1. Obtain historical data of wind turbines and construct a training set based on the results of wind turbine status affected by wake flow; based on AdaB oo st algorithm, combining training sets to build a strong classifier; S2. Use a strong classifier to determine the wind turbines affected by the wake, use the deflection angle of the wake centerline to determine the upstream wind turbines that generate the wake, and define all wind turbines that are not affected by the wake and whose wake does not affect the downstream wind turbines as controllable wind turbine groups; S3. Based on the controllable capacity of each wind turbine in the controllable wind turbine group, a dynamic programming algorithm is used to solve the wind turbine group with the least number of execution actions in the wind farm, and the wind turbines in the group are used as the final execution wind turbines to perform active power control operations based on the controllable capacity of each wind turbine.

2. The method for controlling active power of a wind farm according to claim 1, characterized in that: The historical data of the wind turbine is obtained, and a training set is constructed according to the result of the wind turbine state being affected by the wake; based on AdaB oo The st algorithm, combined with the training set to build a strong classifier, includes the following steps: S 11. Obtain historical data of the wind turbine, and determine whether the state of the wind turbine corresponding to the historical data is affected by the wake according to the change of the deviation between the ambient wind direction and the local wind direction with the wind direction angle, wherein the historical data includes historical operation data, environmental historical data and meteorological historical data; S12, combining historical data and the results of whether the data is affected by wake turbulence to obtain a training set; initializing the weight distribution of each sample in the training set, and adjusting the weight of each sample according to whether the sample classification is correct; S13, based on the weight distribution, establish a weak classifier according to the attribute characteristics of the historical operation data, the environmental historical data and the meteorological historical data, and select the value that minimizes the classification error of the training set as the threshold of the weak classifier; S 14. Based on the weight coefficient formula, the weight of the weak classifier is calculated, and the weight distribution of the training set is updated according to the weight of the weak classifier to obtain a new weight distribution of the sample; S 15. Calculate the judgment value of the overall classification according to the weighted average of each weak classifier, and construct a final strong classifier based on the classification result of the judgment value.

3. The active power control method of a wind farm according to claim 2, characterized in that: The combination of historical data and the result of whether it is affected by wake turbulence obtains a training set; Initialize the weight distribution of each sample in the training set, and adjust the weight of each sample according to the correctness of the sample classification. Combining historical data and the results of whether it is affected by wake turbulence, we get the training set M = {(x1, y1), (x2, y2), ..., (x i ,y i ),…,(x m ,y m )}; Initialize the weight distribution D1 of each sample in the training set = (ω 11 ,ω 12 ,…,ω 1m ),ω 1i =1 / m; Determine whether each sample in the training set is accurately classified. If so, reduce the weight of the sample when constructing the next training set. If not, increase the weight of the sample when constructing the next training set. In the formula, x i represents the i-th historical data, y i represents the wake influence state corresponding to the i-th historical data, m represents the number of samples, D1 represents the weight of the training set at the first iteration, ω 1i Represents the weight of the i-th historical data in the first iteration.

4. The active power control method of a wind farm according to claim 3, characterized in that: The expression of the weak classifier is: h t (x)=ξ(M,D t ) The expression of classification error is: In the formula, h t (x) represents the tth weak classifier, ξ(M, D t ) represents the classifier function, D t represents the weight of the training set at the tth iteration, e t represents the tth weak classifier h t (x) will have weight D t The classification error after the classification of the data set M, ω ti represents the weight of the i-th sample in the t-th iteration, h t (x i ) means that the i-th historical data is classified according to the classifier h t The classification result is obtained by the division method of (x), and I represents the integer weight function.

5. The method for controlling active power of a wind farm according to claim 4, characterized in that: The weight coefficient formula is: The expression for updating the weight distribution of the training set is: D t+1 =(ω (t+1)1， Oh, oh (t+1)2， …, oh (t+1)m ) In the formula, α t Represents the classifier h t (x) Weight coefficient in the final classifier, D t+1 represents the weight of the training set M at the t+1th iteration, ω (t+1)i represents the weight of the i-th historical data in the t+1th iteration, Z t represents the normalization factor.

6. The method for controlling active power of a wind farm according to claim 2, characterized in that: The expression of the judgment value is: The final strong classifier expression is: In the formula, f(x) represents the judgment value of each weak classifier combined with the weight coefficient to classify the historical data, t represents the number of iterations, T represents the total number of iterations, α t Represents the classifier h t (x) is the weight coefficient in the final classifier, h t (x) represents the tth weak classifier, h(x) represents the final strong classifier, and sign(f(x)) represents the judgment value classification function.

7. The method for controlling active power of a wind farm according to claim 1, characterized in that: The method of using a strong classifier to determine the wind turbines affected by the wake, using the wake centerline deflection angle to determine the upstream wind turbines generating the wake, and defining all wind turbines that are not affected by the wake and whose wake does not affect the downstream wind turbines as a controllable wind turbine group includes the following steps: S21, using a strong classifier to determine the wind turbine affected by the wake, and to determine all upstream wind turbines within a preset range upwind of the wind turbine; S22, calculating the wake centerline deflection angles of all upstream wind turbines within a preset range based on the expression of the wake centerline deflection angle; S23, determining whether the wake centerline deflection angle of the upstream wind turbine is equal to the angle difference between the local wind direction of the wind turbine and the ambient wind direction; if so, determining that the upstream wind turbine is an upstream wind turbine that generates wake influence; if not, determining that the upstream wind turbine is not an upstream wind turbine that generates wake influence; S24. All wind turbines that are not affected by wake vortices and whose wake vortices do not affect downstream wind turbines are defined as controllable wind turbine groups.

8. The method for controlling active power of a wind farm according to claim 7, characterized in that: The expression of the wake centerline deflection angle is: Where δ(x) represents the deflection angle of the centerline of the upstream wind turbine wake, c p represents the yaw wind power coefficient of the upstream wind turbine, x represents the distance between the upstream wind turbine and the current wind turbine, k d represents the wake diffusion coefficient, D represents the fan impeller diameter, θ represents the yaw angle of the upstream fan, C t Represents the thrust coefficient of the upstream wind turbine rotor.

9. The method for controlling active power of a wind farm according to claim 1, characterized in that: The method of performing active power control based on the controllable capacity of each wind turbine in the controllable wind turbine group and using a dynamic programming algorithm to solve the wind turbine group with the least number of wind turbines performing actions in the wind farm, and using the wind turbines in the wind turbine group as the final execution wind turbines, comprises the following steps: S31. Obtain the controllable capacity p of each fan in the controllable fan group i , and establish a set of controllable capacities in the controllable wind turbine group, determine the target active power control amount P0; construct two arrays for storing the minimum number of wind turbines d[P] and wind turbine numbers required to achieve each active power control amount; Among them, the expression of the set of controllable capacities is: P 可控 ={p1,p2,...,p i ,...,p n } Where P 可控 represents the set of controllable capacities in the controllable wind turbine group, n represents the number of wind turbines in the controllable wind turbine group, and pi represents the controllable capacity of the i-th wind turbine in the controllable wind turbine group; S32, initializing the minimum number of wind turbines required when the target active power control amount is zero d[0]=0; calculating the minimum number of wind turbines required for each active power control amount based on the state transfer equation; Among them, the expression of the state transfer equation is: Where d[P0] represents the minimum number of wind turbines required for the target active power control quantity P0, d[P0-p i ] indicates that the target active power control amount is P0-p i The minimum number of wind turbines required; S33, gradually increase the active power control amount from zero, calculate the value of the minimum number of wind turbines d[P] required for the active power control amount of each sub-problem, and record the optimal solution of each state and the number of wind turbines that reach the state; S34, iteratively calculate until the target active power control amount P0 is reached, and output the minimum number of wind turbine groups and their numbers required to achieve the target active power control amount P0.

10. An active power control system for a wind farm, used to implement the steps of the active power control method for a wind farm as claimed in any one of claims 1 to 9, characterized in that: The active power control system of the wind farm includes a strong classifier building module, a controllable wind turbine group determination module and a dynamic programming solution module; The strong classifier building module is used to obtain the historical data of the wind turbine and build a training set based on the results of the wind turbine status being affected by the wake. oo st algorithm, combining training sets to build a strong classifier; The controllable wind turbine group determination module is used to determine the wind turbines affected by the wake using a strong classifier, determine the upstream wind turbines generating the wake using the wake centerline deflection angle, and define all wind turbines that are not affected by the wake and whose wake does not affect the downstream wind turbines as the controllable wind turbine group; The dynamic programming solution module is used to solve the wind turbine group with the least number of execution actions in the wind farm based on the controllable capacity of each wind turbine in the controllable wind turbine group using a dynamic programming algorithm, and use the wind turbines in the group as the final execution wind turbines to perform active power control operations based on the controllable capacity of each wind turbine.

Images