Maximum power capture method, system and application of variable speed wind turbine generator system

Abstract

A variable speed wind turbine maximum power capture method, system and application thereof, the method is based on active disturbance rejection control, RBF neural network and variable speed wind turbine maximum power capture method, including the following steps: establishing the equivalent model of variable speed wind turbine; using the second order linear active disturbance rejection control (LADRC) algorithm to construct the generator torque controller; using the gravity search algorithm to optimize the generator torque controller parameters and obtain the optimal parameter data set; based on the optimal parameter data set, the radial basis function neural network is trained, and the trained radial basis function neural output is suitable for the generator torque controller parameters under the current wind speed condition. Using the gravity search algorithm to optimize the generator torque controller parameters and obtain the optimal parameter data set; based on the optimal parameter data set, the radial basis function neural network is trained, and the trained radial basis function neural output is suitable for the generator torque controller parameters under the current wind speed condition.

Description

Technical Field

[0001] The present invention relates to maximum power point tracking control of a wind turbine generator set, and more particularly to a maximum power capture method and system for a variable speed wind turbine generator set and applications thereof. Background Art

[0002] Among renewable energy sources, the use of wind energy is rapidly expanding due to its economy and safety. For variable-speed wind turbines, generator torque control is usually used to maximize the power captured from the wind when the wind speed is lower than the rated wind speed. Feasibility studies on the construction of wind power stations have shown that in many areas of the earth, the wind speed is mostly low throughout the year (Mostafaeipour A. Feasibility study of offshore wind turbine installation in Iran compared with the world [J]. Renewable and Sustainable Energy Reviews, 2010, 14 (7): 1722-1743.). Therefore, in order to maximize the achievable power, generator torque control plays a more important role in VSWT. The literature (Amine HM, Abdelaziz H, Najib E. Wind turbine maximum power point tracking using FLC tuned with GA [J]. Energy Procedia, 2014, 62: 364-373.) uses a fuzzy logic controller based on a genetic algorithm to track the maximum power point of a wind turbine; the literature (Mérida J, Aguilar LT, Dávila J. Analysis and synthesis of sliding mode control for large scale variable speed wind turbine for power optimization [J]. Renewable Energy, 2014, 71: 715-728.) uses the synthesis and analysis of sliding mode control for power optimization of large scale variable speed wind turbines. The paper (Boukhezzar B, Siguerdidjane H. Nonlinear control with wind estimation of a DFTG variable speed wind turbine for power capture optimization [J]. Energy Conversion and Management, 2009, 50(4): 885-892) employs nonlinear control and uses wind force approximation of a DFTG variable speed wind turbine to optimize power capture while avoiding strong transients in turbine components, particularly the drive train. Existing technologies do not handle wind turbine model uncertainty and unmodeled dynamics well, and are significantly affected by wind speed.Therefore, the application proposes a wind turbine maximum power point tracking strategy based on active disturbance rejection control and neural network.

[0003] Neural networks are a powerful method for approximating arbitrary input-output mappings. Radial-basis function networks are a successful example of this approach. The structure of a radial-basis function (RBF) neural network includes a single hidden layer of non-linear nodes, with center locations that cause each node to specialize in a particular region of the input space. The desired response is obtained through a training procedure that adjusts the weights connecting the hidden layer to linear output nodes. The main advantages of neural networks are: (1) the neural weights are adjusted online, without any pre-training phase; (2) the stability and performance of the closed-loop system can be guaranteed. Therefore, neural network control is very suitable for controlling uncertain nonlinear dynamic systems. SUMMARY

[0004] In order to solve the above-mentioned deficiencies in the prior art, the application discloses a variable-speed wind turbine maximum power capture method, and the technical scheme is as follows:

[0005] A variable-speed wind turbine maximum power capture method, which is a variable-speed wind turbine maximum power capture method based on active disturbance rejection control, RBF neural network and gravitational search algorithm, characterized by comprising the following steps:

[0006] Step 1: establishing an equivalent model of a variable-speed wind turbine;

[0007] Step 2: using a second-order linear active disturbance rejection control (LADRC) algorithm to construct a generator torque controller;

[0008] Step 3: using a gravitational search algorithm to optimize the generator torque controller parameters and obtain an optimal parameter data set;

[0009] Step 4: training a radial-basis function neural network based on the optimal parameter data set, and using the trained radial-basis function neural output to adapt the generator torque controller parameters to the current wind speed conditions.

[0010] The application further discloses a variable-speed wind turbine maximum power capture method applied to a variable-speed wind turbine control system.

[0011] Advantages:

[0012] The proposed scheme can accurately track the maximum power point and still exhibit good performance under turbulent wind conditions. BRIEF DESCRIPTION OF DRAWINGS

[0013] Figure 1The schematic diagram of the variable speed wind turbine maximum power capture method based on active disturbance rejection control, RBF neural network and gravitational search algorithm. DETAILED DESCRIPTION

[0014] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the embodiments in the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all the other embodiments obtained by those skilled in the art without creative work belong to the protection scope of the present application.

[0015] The variable speed wind turbine maximum power capture method is based on active disturbance rejection control, RBF neural network and gravitational search algorithm, and characterized by comprising the following steps:

[0016] Step 1: establishing an equivalent model of the variable speed wind turbine;

[0017] Step 2: adopting a second-order linear active disturbance rejection control (LADRC) algorithm to construct a generator torque controller;

[0018] Step 3: adopting a gravitational search algorithm to optimize the generator torque controller parameters and obtain an optimal parameter data set;

[0019] Step 4: training a radial basis function neural network based on the optimal parameter data set, and adopting the trained radial basis function neural output to obtain the generator torque controller parameters suitable for the current wind speed condition.

[0020] Next, how to realize each step will be described in detail, so as to finally solve the technical problems to be solved by the present application and achieve the technical effects.

[0021] Step 1: establishing an equivalent model of the variable speed wind turbine

[0022] The aerodynamic power captured by the wind turbine can be described as

[0023]

[0024] where ρ is the air density, R is the rotor radius, v is the wind speed, C p is the power coefficient of the wind turbine. C p depends on the pitch angle β and the tip speed ratio λ, which is defined as follows:

[0025]

[0026] where ω is the rotor speed. The aerodynamic torque T aero is represented as follows:

[0027]

[0028] where C q is the torque coefficient, and C p is the relationship as follows:

[0029]

[0030] The present application uses a simple first-order generator model, therefore, the generator torque T gen can be expressed as:

[0031]

[0032] where T ref is the generator torque reference value, τ gen is the time constant of the generator. The generator power P gen is determined by the following formula:

[0033] P gen = T gen ω gen (40)

[0034] where ω gen is the generator speed.

[0035] The rotor inertia J r is driven by the aerodynamic torque T aero at the speed ω r , this process can be described by the following dynamics equation:

[0036]

[0037] where K r is the external damping of the rotor. The low-speed shaft torque T ls acts as a braking torque on the rotor, which is obtained by the following equation:

[0038] T ls = K ls (φ r -φ ls ) + B ls (ω r -ω ls ) (42)

[0039] where is the angular deviation on the rotor side, is the angular deviation on the gearbox side. ω ls is the speed of the low-speed shaft. In addition, B ls and K ls are the low-speed shaft stiffness and low-speed shaft damping, respectively. The generator inertia is driven by the high-speed shaft, and is braked by the generator torque:

[0040]

[0041] The gearbox speed ratio is defined as:

[0042]

[0043] In the maximum power capture region of wind turbine operation, the wind speed is higher than the cut-in wind speed but lower than the rated wind speed. The main control objective in this region is to maximize the power generated by the wind turbine. In order to capture the maximum power, the power coefficient C p Should be set to the maximum power factor C p，max , at this time the tip speed ratio and pitch angle remain at the optimal value:

[0044]

[0045] In this region, P gen It can be described as:

[0046]

[0047] In the rated power area, the wind speed is higher than the rated wind speed but lower than the cut-out wind speed. The main control purpose of this area is to make the generator power P gen Maintain the rated generator power P gen，n To achieve this goal, T ref Defined as:

[0048]

[0049] Step 2: Use the second-order linear active disturbance rejection control (LADRC) algorithm to design the generator torque controller. In the design process of the second-order LADRC control system, it is assumed that the mathematical model of the controlled object is

[0050]

[0051] Where b0 is the high-frequency gain of the controlled object, and f(y, u, d) includes the internal uncertainty of the object and the external disturbance, that is, the total disturbance. Let z = [z1 z2 z3] T , where z1 = y, z3=f, model (48) can be expressed as

[0052]

[0053] in,

[0054]

[0055] Design a linear extended state observer for equation (50):

[0056]

[0057] in, is the estimated value of z, L o is the observer gain, we have

[0058] L o =[β1 β2 β3] T (52)

[0059] When A e -L o C e When the process becomes stable, The total disturbance can be estimated. In order to eliminate the total disturbance through the feedback loop, select

[0060]

[0061] Among them, K o is the feedback control gain, which can be expressed as

[0062] K o =[k p k d 1] / b0 (54)

[0063] is the expanded reference input signal, which can be expressed as

[0064]

[0065] The parameter that needs to be tuned for the second-order LADRC is the feedback controller gain K o and the ESO observer gain L o To facilitate tuning, the literature (Gao Z. Scaling and bandwidth-parameterization based controller tuning [C] / / Proceedings of the American control conference. 2006, 6: 4989-4996.) proposed a bandwidth tuning method, which only requires adjusting the controller bandwidth ω c , observer bandwidth ω o The three parameters b0 and b1 can achieve satisfactory controller performance. The specific method is: In equations (52) and (54), by adjusting the controller bandwidth ω c and observer bandwidth ω o Get the gain K o , L o The various elements of

[0066]

[0067] Step 3: Use the gravitational search algorithm to optimize the generator torque controller parameters and obtain the optimal parameter data set.

[0068] Newton's laws of motion and gravity are used to achieve the optimal solution. Assume a system of masses, where the position of the i-th mass is described as follows:

[0069]

[0070] in Represents the position of the i-th mass in the d-th dimension, and n represents the dimension of the search space. The mass of each mass body is calculated based on the fitness value of the current population:

[0071]

[0072] Among them, M i (t) and fit i (t) are the mass and fitness value of the i-th mass body at the t-th moment. The Worst(t) and Best(t) of the minimization problem are described as follows:

[0073]

[0074] Use the law of universal gravitation to calculate the acceleration of a mass:

[0075]

[0076] Then update the velocity and position of the mass body according to the following formula:

[0077]

[0078] where rand i and rand j Represents two uniform random numbers distributed in the interval [0, 1]. ε is a very small value. R ij (t) represents the Euclidean distance between the i-th mass body and the j-th mass body. kbest represents the set of the first k mass bodies with the largest mass, which is a function of time, initialized to K0 at the beginning and reduced over time. Here K0 is adjusted to s (the total number of mass bodies) and is linearly reduced to 1. In the gravitational search algorithm, the gravitational constant G will take an initial value G0 and decrease over time:

[0079] G(t)=G(G0,t) (62)

[0080] The gravitational search algorithm consists of the following stages:

[0081] (a) Determine the search space and initialize it randomly.

[0082] (b) Calculate the fitness value fit of the mass body i (t).

[0083] (c) Update G(t), best(t), worst(t) and M according to formulas (62), (59) and (58) respectively. i (t).

[0084] (d) Calculate the total force and acceleration in different directions according to formula (60).

[0085] (e) Update the position and velocity of the mass body according to formula (61).

[0086] (f) Repeat steps b to e until the stopping criterion is reached.

[0087] Step 4: Train the radial basis function neural network based on the optimal parameter data set, and use the trained radial basis function neural network to output the generator torque controller parameters suitable for the current wind speed conditions.

[0088] The radial basis function neural network uses the radial basis function as the activation function, which is a linear combination of radial basis functions. RBF neural networks usually have three layers: an input layer, a hidden layer with a nonlinear RBF activation function, and a linear output layer. X = [x1, x2, ..., x n ] T is the input vector, is the radial basis vector, Y = [y1, y2, ..., y m ] T Is the output vector. Hidden layer (ψ j ) is a Gaussian function that can be described as

[0089]

[0090] Among them, C j =[c j1 , c j2 ,...,c jn ] T Is the center vector of the jth neuron. σ=[σ1,σ2,…,σ p ] T is the basis width vector which can usually be determined experimentally. XC j Yes (XC j ) can be described by the following formula:

[0091]

[0092] The i-th output of the RBF neural network is given by:

[0093]

[0094] Where W iq is the weight of the qth hidden neuron on the ith output.

[0095] The training process of the RBF neural network is carried out through the following algorithm:

[0096] Step 1: Initialize all weights randomly.

[0097] Step 2: Calculate each element of the output vector Y through (65).

[0098] Step 3: Calculate the error ε of each neuron in the output layer by the following formula i :

[0099]

[0100] in is the expected output of the i-th neuron in the output layer.

[0101] Step 4: Update the weights by the following equation:

[0102] W ij (n+1)=W ij (n)+με i (n)ψ j ,i=1,2,...,m,j=1,2,...,p (67)

[0103] Where n is the number of iterations and μ is the learning rate.

[0104] Step 5: Calculate the total error ε according to the following formula T

[0105]

[0106] Step 6: Return to step 3 and repeat the calculation until ε T Smaller than the expected error.

[0107] 1. Solution proposed by the present invention

[0108] First, choose the fitness function:

[0109]

[0110] The gravitational search algorithm is used to find the linear active disturbance rejection controller parameters b0, ω o ,ω cThe optimization is performed at different wind speed points to obtain the optimal controller parameter training set. Then, this optimal controller parameter training set and wind speed are used as input data for the RBF neural network to train the neural network. The RBF neural network can then output the optimal parameters suitable for the current wind speed.

[0111] Example 1

[0112] The wind turbine of the present invention is a 5MW horizontal-axis variable-speed wind turbine. The technical parameters of the turbine are detailed in the literature (Jonkman J, Butterfield S, Musial W, et al. Definition of a 5-MW reference wind turbine for offshore system development [R]. National Renewable Energy Lab. (NREL), Golden, CO (United States), 2009). Some of the parameters are shown in Table 1.

[0113] Table 1 Parameters of 5MW horizontal axis variable speed wind turbine

[0114]

[0115] The C of the wind turbine p，max , β opt and λ opt The values ​​of are 0.482, 0° and 7.55 respectively. The air density ρ of the wind turbine studied in this paper is 1.02 (kg / m3), and K is calculated opt It is 2.853044.

[0116] First, the gravitational search algorithm is used to find the linear active disturbance rejection controller parameters b0, ω o ,ω c Optimization is performed at different wind speed points to obtain the best controller parameter training set. The specific optimization process is as follows:

[0117] Assume a system with masses, where the position of the i-th mass is described as follows:

[0118]

[0119] in Represents the position of the i-th mass in the d-th dimension, and n represents the dimension of the search space. The mass of each mass body is calculated based on the fitness value of the current population:

[0120]

[0121] Among them, M i (t) and fiti (t) are the mass and fitness value of the i-th mass body at the t-th moment. The Worst(t) and Best(t) of the minimization problem are described as follows:

[0122]

[0123] Use the law of universal gravitation to calculate the acceleration of a mass:

[0124]

[0125] Then update the velocity and position of the mass body according to the following formula:

[0126]

[0127] where rand i and rand j Represents two uniform random numbers distributed in the interval [0, 1]. ε is a very small value. R ij (t) represents the Euclidean distance between the i-th mass body and the j-th mass body. kbest represents the set of the first k mass bodies with the largest mass, which is a function of time, initialized to K0 at the beginning and reduced over time. Here K0 is adjusted to s (the total number of mass bodies) and is linearly reduced to 1. In the gravitational search algorithm, the gravitational constant G will take an initial value G0 and decrease over time:

[0128] G(t)=G(G0,t) (75)

[0129] The gravitational search algorithm consists of the following stages:

[0130] (a) Determine the search space and initialize it randomly.

[0131] (b) Calculate the fitness value of the mass body according to formula (69): i (t).

[0132] (c) Update G(t), best(t), worst(t) and M according to formulas (75), (72) and (71) respectively. i (t).

[0133] (d) Calculate the total force and acceleration in different directions according to formula (73).

[0134] (e) Update the position and velocity of the mass body according to formula (74).

[0135] (f) Repeat steps b to e until the stopping criterion is reached.

[0136] Then, the obtained optimal controller parameter training set and wind speed are used as input data for the RBF neural network to train the neural network. The radial basis function neural network uses the radial basis function as the activation function, which is a linear combination of radial basis functions. The RBF neural network usually has three layers: an input layer, a hidden layer with a nonlinear RBF activation function, and a linear output layer. X = [x1, x2, ..., x n ] T is the input vector, is the radial basis vector, Y = [y1, y2, ..., y m ] T Is the output vector. Hidden layer (ψ j ) is a Gaussian function that can be described as

[0137]

[0138] Among them, C j =[c j1 , c j2 ,...,c jn ] T Is the center vector of the jth neuron. σ=[σ1,σ2,…,σ p ] T is the basis width vector which can usually be determined experimentally. XC j Yes (XC j ) can be described by the following formula:

[0139]

[0140] The i-th output of the RBF neural network is given by:

[0141]

[0142] Where W iq is the weight of the qth hidden neuron on the ith output.

[0143] The training process of the RBF neural network is carried out through the following algorithm:

[0144] Step 1: Initialize all weights randomly.

[0145] Step 2: Calculate each element of the output vector Y using formula (78).

[0146] Step 3: Calculate the error ε of each neuron in the output layer by the following formula i :

[0147]

[0148] in is the expected output of the i-th neuron in the output layer.

[0149] Step 4: Update the weights by the following equation:

[0150] W ij (n+1)=W ij (n)+με i (n)ψ j ,i=1,2,...,m,j=1,2,...,p (80)

[0151] Where n is the number of iterations and μ is the learning rate.

[0152] Step 5: Calculate the total error ε according to the following formula T

[0153]

[0154] Step 6: Return to step 3 and repeat the calculation until ε T Smaller than the expected error.

[0155] Finally, the RBF neural network can output the optimal controller parameters suitable for the current wind speed, so that the variable-speed wind turbine can achieve maximum power capture.

[0156] In summary, the proposed solution accurately tracks the maximum power point and exhibits good performance even in turbulent wind conditions. This is because the linear active disturbance rejection controller considers unmodeled dynamics and model uncertainty as part of the total system disturbance, estimates them using the extended state observer, and compensates for them using the feedback control law. Furthermore, the RBF neural network outputs optimal controller parameters based on the current wind speed, enabling the controller to maximize its performance.

[0157] In the above description, many specific details are set forth in order to fully understand the present invention. However, the above description is only a preferred embodiment of the present invention. The present invention can be implemented in many other ways different from those described herein, so the present invention is not limited to the specific implementation disclosed above. At the same time, any person skilled in the art can make many possible changes and modifications to the technical solution of the present invention using the methods and technical contents disclosed above without departing from the scope of the technical solution of the present invention, or modify it into an equivalent embodiment of equivalent changes. Any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention still falls within the scope of protection of the technical solution of the present invention.

Claims

1. A method for capturing maximum power of a variable-speed wind turbine generator set, which is based on active disturbance rejection control, radial basis function (RBF) neural network, and gravitational search algorithm, and is characterized by: The steps include: Step 1: Establish an equivalent model of the variable speed wind turbine; Step 2: Use the second-order linear active disturbance rejection control (LADRC) algorithm to build the generator torque controller; Step 3: Use the gravitational search algorithm to optimize the generator torque controller parameters and obtain the optimal parameter data set; Step 4: Train the radial basis function (RBF) neural network based on the optimal parameter data set. Use the trained radial basis function (RBF) neural network to output generator torque controller parameters suitable for the current wind speed conditions, so that the variable speed wind turbine can achieve maximum power capture.

2. A maximum power capture system for a variable speed wind turbine generator set, the system adopting the maximum power capture method for a variable speed wind turbine generator set according to claim 1.

3. A variable speed wind turbine generator control system, characterized in that: The maximum power capture method of a variable-speed wind turbine generator set according to claim 1 is applied to a control system of a variable-speed wind turbine generator set.

Images