A coordinated optimization control method and device for improving active regulation capability of a wind turbine generator system

Abstract

The application discloses a kind of coordinated optimization control method and device for improving the active regulation ability of wind turbine, and relates to wind turbine control technical field.The method includes constructing first-order error active disturbance rejection controller;Establish the controller parameter optimization model with the minimum fatigue load of unit and the best control performance as multi-objective;According to the operating parameters of wind turbine, obtain the fatigue load index value using bidirectional long short-term memory network;According to the fatigue load index value, the output power and rotor speed of wind turbine, the controller parameter optimization model is solved using snow melting optimization algorithm, and the optimal parameters of two first-order error active disturbance rejection controllers are obtained;The pitch angle and electromagnetic torque of wind turbine are controlled respectively by two first-order error active disturbance rejection controllers using optimal parameters, to adjust the output power and rotor speed of wind turbine.The application can improve the active instruction tracking ability of wind turbine, while reducing fatigue load.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of wind turbine control, in particular to a coordinated optimization control method and device for improving the active regulation capability of a wind turbine. BACKGROUND

[0002] With the rapid development of renewable energy, wind energy as a clean and sustainable energy form is increasingly valued by people. As a primary conversion device for wind energy, the stable operation and long-term reliability of wind turbine generators are crucial to the development of the entire wind power industry. Traditional wind turbine control is designed to output maximum power while operating safely. However, as the penetration of wind power in the power grid increases, the demand for active power control provided by wind power is increasingly evident. In addition, when wind power is connected to the grid, wind turbine generators should have the ability to track active power commands. However, this also causes the fatigue load of the wind turbine to increase significantly. Therefore, in the active power control of wind turbine generators, wind energy capture efficiency is no longer the core concern.

[0003] In research and engineering applications, fatigue load is mainly reflected in the unbalanced torque borne by the main shaft and the cyclic bending moment generated by the tower under the action of wind. Therefore, the standard deviation of the main shaft torque and the standard deviation of the tower bending moment are widely considered as typical fatigue indicators of wind turbines. However, the calculation of main shaft torque and tower bending moment is complex, with numerous influencing parameters, making it difficult to calculate accurate results in practical applications.

[0004] In order to realize the active power control of wind turbines, a controller is needed to manipulate the reference values of the pitch angle and generator torque. The main goal of controller design is to achieve fast and accurate tracking of active power commands through coordinated pitch angle and torque control. In the design of wind turbine control strategies, wind speed and power command fluctuations are considered as external disturbances, and unmodeled parts and modeling uncertainties are usually considered as internal disturbances. Active Disturbance Rejection Control (ADRC) has excellent performance in resisting external and internal disturbances. However, due to the multiple parameters and double freedom characteristics of ADRC, it is difficult to be widely promoted in industry. And the active power control of wind power aggravates the fatigue load of the unit, resulting in higher component failure rate and maintenance cost. Therefore, how to improve the active command tracking capability of wind turbine generators and reduce the fatigue load of the system has become the main target that needs to be solved urgently at present. SUMMARY

[0005] The purpose of the present application is to provide a coordinated optimization control method and device for improving the active regulation capability of a wind turbine, which can improve the active command tracking capability of a wind turbine generator while reducing fatigue load.

[0006] To achieve the above object, the application provides the following scheme.

[0007] In a first aspect, the application provides a coordinated optimization control method for improving the active regulation capability of a wind turbine, comprising: constructing a first-order error active disturbance rejection controller; establishing a controller parameter optimization model with multiple objectives of minimizing the fatigue load of the wind turbine and optimizing the control performance; the controller parameter optimization model including parameters of two first-order error active disturbance rejection controllers; obtaining a fatigue load index value by using a bidirectional long short-term memory network according to the operating parameters of the wind turbine; solving the controller parameter optimization model by using a snow melt optimization algorithm according to the fatigue load index value, the output power and the rotor speed of the wind turbine, to obtain optimal parameters of the two first-order error active disturbance rejection controllers; adjusting the output power and the rotor speed of the wind turbine by controlling the pitch angle and the electromagnetic torque of the wind turbine respectively through the two first-order error active disturbance rejection controllers using the optimal parameters.

[0008] Optionally, the first-order error active disturbance rejection controller comprises: an extended state observer and a control law; the extended state observer takes the system error and the control signal output by the control law as inputs, and outputs first and second state variables according to the system error and the control signal output by the control law; when the system error is the deviation of the output power of the wind turbine from the active power reference value, the control signal output by the control law is the pitch angle command; when the system error is the deviation of the rotor speed of the wind turbine from the rotor speed reference value, the control signal output by the control law is the electromagnetic torque command; the gain of the extended state observer satisfies l1=2ω o , l1 and l2 are the gains of the extended state observer, ω o is the bandwidth of the extended state observer; the control law outputs the control signal according to the first and second state variables according to ; wherein u is the control signal, k0 and b0 are control gains, k0=ω c , ω c is the bandwidth of the first-order error active disturbance rejection controller, z1 is the first state variable, and z2 is the second state variable.

[0009] Optionally, the controller parameter optimization model is:

[0010]

[0011] wherein, is a parameter matrix of the first-order error active disturbance rejection controller, b 01 and b 02ω o1 and ω o2 are the bandwidths of the extended state observer of the two first-order error active disturbance rejection controllers, ω c1 and ω c2 are the bandwidths of the two first-order error active disturbance rejection controllers; F is a total target function; F1 is a control performance sub-target function; F2 is a fatigue load sub-target function; h1 is a weighting coefficient of the control performance sub-target function; h2 is a weighting coefficient of the fatigue load sub-target function; T0 is a preset initial time; T max is a maximum simulation time; t is a time variable; is a rotor speed reference value; ω r is a rotor speed of the wind turbine; P cmd is an active power reference value; P e is an output power of the wind turbine; std is a standard deviation; M t is a tower bending moment; T s is a main shaft torque; β ref and Δβ ref are respectively a pitch angle instruction and a pitch angle instruction change amount; T g-ref and ΔT g-ref are respectively an electromagnetic torque instruction and an electromagnetic torque instruction change amount; s is a time unit; Z max and Z min are respectively an upper limit and a lower limit of a parameter variable of the first-order error active disturbance rejection controller.

[0012] Optionally, according to the operating parameters of the wind turbine, the fatigue load index value is obtained by using a bidirectional long short-term memory network, specifically including: determining the tower bending moment and the main shaft torque as the fatigue load index; obtaining a plurality of operating parameter data of the wind turbine; according to the plurality of operating parameter data, performing Pearson correlation analysis on the plurality of operating parameters and the tower bending moment and the main shaft torque to obtain a Pearson correlation coefficient of each operating parameter with the tower bending moment and a Pearson correlation coefficient of each operating parameter with the main shaft torque; selecting an operating parameter with a Pearson correlation coefficient greater than a preset coefficient threshold as a first input characteristic variable corresponding to the tower bending moment and a second input characteristic variable corresponding to the main shaft torque; according to the first input characteristic variable value of the wind turbine, using a first bidirectional long short-term memory network to obtain a numerical value of the tower bending moment; according to the second input characteristic variable of the wind turbine, using a second bidirectional long short-term memory network to obtain a numerical value of the main shaft torque.

[0013] Optionally, the first input characteristic variable corresponding to the tower bending moment includes a pitch angle, a wind speed, a mechanical torque, and a tip speed ratio; and the second input characteristic variable corresponding to the main shaft torque includes a rotor speed, an electromagnetic speed, a mechanical torque, and an electromagnetic torque.

[0014] Optionally, the specific process of the snow melt optimization algorithm comprises: initializing population parameters; the population parameters comprise an initial population size, a maximum iteration number and a search range; initializing initial positions of all individuals in the initial population using a Sobol sequence; calculating fitness values of each individual in the initial population according to the initial positions of each individual, and selecting an initial position of an individual with the largest fitness value as a global optimal solution; constructing an elite pool of the initial population; introducing a double population mechanism to divide the entire initial population into two sub-populations; entering an exploration stage, simulating a complex process of snow or water being converted into vapor using Brownian motion, and updating the two sub-populations; entering a development stage, simulating a process of snow melting into water using a degree-day method; updating positions of each individual in the current entire population using a golden sine search mechanism; calculating fitness values of each individual in the current entire population, and updating the global optimal solution; mutating the global optimal solution using a Levy flight strategy with long-short jump capability to generate a mutated global optimal solution; determining whether the current global optimal solution is the global optimal solution before mutation or the mutated global optimal solution using a greedy strategy; updating the elite pool; determining whether the current iteration count has reached the maximum iteration number, obtaining a comparison result; if the comparison result is no, returning to the step of introducing the double population mechanism to divide the entire initial population into two sub-populations; and if the comparison result is yes, outputting the current global optimal solution.

[0015] Optionally, the elite pool is [Z b est , Z second (k), Z third (k), Z c (k)]; wherein, Z b est is a global optimal solution of the population, Z second (k) and Z third (k) are positions of the second best and third best individuals in the population, and Z c (k) is a center position of the top 50% individuals with fitness values. is a total population after rearranging Z i (k) from best to worst, and N1 is half of the total population size; k = 1, 2, …, K, and K is a maximum iteration number.

[0016] Optionally, the greedy strategy is: wherein, Z b ′ est is the current global optimal solution, Z new is the mutated global optimal solution, Z best is the global optimal solution before mutation, and f is a fitness function.

[0017] ​​Optionally, the pitch angle and the electromagnetic torque of the wind turbine are controlled by two first-order error active disturbance rejection controllers using optimal parameters, specifically including: inputting a deviation between the output power of the wind turbine and an active power reference value into a first first-order error active disturbance rejection controller to output a pitch angle instruction; the optimal parameters of the first first-order error active disturbance rejection controller include b 01 , ω o1 and ω c1 ; inputting a deviation between the rotor speed of the wind turbine and a rotor speed reference value into a second first-order error active disturbance rejection controller to output an electromagnetic torque instruction; the optimal parameters of the second first-order error active disturbance rejection controller include b 02 , ω o2 and ω c2 .

[0018] In a second aspect, the application provides a coordinated optimization control device for improving the active regulation capability of a wind turbine, including: a controller construction module, an optimization model establishment module, an index calculation module, a parameter optimization module and a coordinated optimization module.

[0019] The controller construction module is configured to construct a first-order error active disturbance rejection controller; the optimization model establishment module is configured to establish a controller parameter optimization model with multiple targets of minimizing the fatigue load of the wind turbine and optimizing the control performance; the controller parameter optimization model includes parameters of two first-order error active disturbance rejection controllers; the index calculation module is configured to obtain a fatigue load index value by using a bidirectional long short-term memory network according to the operating parameters of the wind turbine; the parameter optimization module is configured to solve the controller parameter optimization model by using a snow melt optimization algorithm according to the fatigue load index value, the output power and the rotor speed of the wind turbine, to obtain optimal parameters of the two first-order error active disturbance rejection controllers; the snow melt optimization algorithm uses Sobol sequence to initialize a population, uses a golden sine search mechanism to update the entire population, and uses a Levy flight strategy to mutate optimal individuals; and the coordinated optimization module is configured to control the pitch angle and the electromagnetic torque of the wind turbine by the two first-order error active disturbance rejection controllers using the optimal parameters, to adjust the output power and the rotor speed of the wind turbine.

[0020] According to the specific embodiments provided by the application, the application has the following technical effects:

[0021] The application provides a coordinated optimization control method and device for improving active regulation capacity of a wind turbine, and when parameters of a first-order error active disturbance rejection controller are optimized, a controller parameter optimization model with multiple targets of minimum fatigue load of the unit and optimal control performance is established, the tracking performance of a power instruction and the reduction of a thermal power unit fatigue load are considered, a snow melt optimization algorithm is used to quickly and accurately locate optimal parameters of the first-order error active disturbance rejection controller, and then two first-order error active disturbance rejection controllers using the optimal parameters are used to control a pitch angle and an electromagnetic torque of the wind turbine respectively, active power and rotor speed output by the wind turbine can be adjusted, the active instruction tracking capacity of the wind turbine is improved, and the fatigue load is reduced. BRIEF DESCRIPTION OF DRAWINGS

[0022] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related art, the drawings needed to be used in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description only constitute some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor on the basis of these drawings.

[0023] Figure 1 A flowchart of a coordinated optimization control method for improving active regulation capacity of a wind turbine is provided for an embodiment of the present application.

[0024] Figure 2 A wind turbine structure diagram is provided for another embodiment of the present application.

[0025] Figure 3 A first-order error active disturbance rejection controller structure diagram is provided for another embodiment of the present application.

[0026] Figure 4 A wind turbine active power coordinated control block diagram based on a first-order error active disturbance rejection controller is provided for another embodiment of the present application.

[0027] Figure 5 A long short-term memory network structure diagram is provided for another embodiment of the present application.

[0028] Figure 6 A bidirectional long short-term memory network structure diagram is provided for another embodiment of the present application.

[0029] Figure 7 A flowchart of an enhanced snow melt optimization algorithm is provided for another embodiment of the present application.

[0030] Figure 8 A principle diagram of a coordinated optimization control method for improving active regulation capacity of a wind turbine is provided for an embodiment of the present application. DETAILED DESCRIPTION

[0031] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0032] The above purposes, features and advantages of the present application will be more apparent and understandable. The present application will be further described in detail below with reference to the drawings and specific embodiments.

[0033] In an exemplary embodiment, as shown in Figure 1 A coordinated optimization control method for improving the active regulation capability of a wind turbine is provided, including the following steps 101 to 105. Among them:

[0034] Step 101: Construct a first-order error active disturbance rejection controller.

[0035] Step 102: Establish a controller parameter optimization model with the minimum fatigue load of the wind turbine and the best control performance as the multi-objective; the controller parameter optimization model includes the parameters of two first-order error active disturbance rejection controllers.

[0036] Step 103: According to the operating parameters of the wind turbine, obtain the fatigue load index value by using a bidirectional long short-term memory network.

[0037] Step 104: According to the fatigue load index value, the output power and the rotor speed of the wind turbine, solve the controller parameter optimization model by using a snow melt optimization algorithm to obtain the optimal parameters of the two first-order error active disturbance rejection controllers; the snow melt optimization algorithm uses Sobol sequence to initialize the population, uses golden sine search mechanism to update the entire population, and uses Levy flight strategy to mutate the optimal individual.

[0038] Step 105: Adjust the output power and the rotor speed of the wind turbine by controlling the pitch angle and the electromagnetic torque of the wind turbine respectively through the two first-order error active disturbance rejection controllers using the optimal parameters.

[0039] Implementing the above steps 101 to 105 not only improves the active instruction tracking capability of the wind turbine, but also reduces the fatigue load of the system.

[0040] In another exemplary embodiment of the present application, in order to be compatible with standard industrial control function blocks, the two-degree-of-freedom ADRC is reconfigured into a one-degree-of-freedom error active disturbance rejection control (EADRC). In the EADRC, the observation gain and the control gain are bandwidthed, which simplifies the problem of controller structure and parameter tuning. In particular, using the error as the input of the controller makes the EADRC more convenient for application and deployment in industry. Therefore, it is a good attempt to introduce the EADRC into the active power control of a wind turbine. Therefore, in order to improve the regulation performance of the active power of the wind turbine, the present application designs a one-order error active disturbance rejection controller according to the model of the wind power system.

[0041] A conventional wind power system is composed of many interrelated subsystems, mainly composed of an aerodynamics subsystem, a transmission subsystem, a variable pitch subsystem, an electrical subsystem and a control subsystem. The structure of the wind turbine is shown in Figure 2 .

[0042] In the wind power control system, the operation of the system can be regarded as a two-order dynamic process, which is generally described as:

[0043]

[0044] In the formula, a0 and a1 are system parameters. u(t) is the input of the system, that is, the control amount; y(t) is the output of the system; and are the state variables of the system; b is the control gain; w(t) is an unknown disturbance.

[0045] The EADRC is a practical one-degree-of-freedom controller, and the structure of the one-order EADRC system is shown in Figure 3 . Figure 3 In the formula, r represents the set value, ε is the system error, y is the output of the system, k0 and b0 are the control gains, u0 is the error feedback control amount, u is the control signal applied to the system after adjustment, and z1 and z2 are the outputs of the extended state observer (ESO).

[0046] The extended state observer takes the system error and the control signal output by the control law as the input, and outputs the first state variable and the second state variable according to the system error and the control signal output by the control law; when the system error is the deviation of the output power of the wind turbine from the active power reference value, the control signal output by the control law is the pitch angle command; when the system error is the deviation of the rotor speed of the wind turbine from the rotor speed reference value, the control signal output by the control law is the electromagnetic torque command.

[0047] According to the characteristics of ADRC, all the dynamic characteristics different from the series integrator can be regarded as disturbances, which are then estimated and compensated by ESO. Considering that low-order controllers are easier to design and deploy in practical applications, while they can stably control high-order objects. A first-order EADRC is designed to control the wind power system in this paper, which simplifies the complexity of controller structure and parameter adjustment. Based on this analysis, the structure of the wind power system can be rewritten as:

[0048]

[0049] Let

[0050]

[0051] Equation (2) can be rewritten as:

[0052]

[0053] In the equation, f is the total disturbance of the system.

[0054] In the framework of the EADRC method, the error is defined as the state variable:

[0055]

[0056] In the equation, h is the total disturbance of the system in the EADRC scheme,

[0057] The state variable is chosen as x = [x1 x2] = [ε h]. Then, equation (5) is written as:

[0058]

[0059] The second-order ESO is given by:

[0060]

[0061] In the equation, z1 and z2 are the estimates of x1 and h, respectively; l1 and l2 are the observer gains.

[0062] The design of the EADRC control method is consistent with the traditional ADRC. In order to simplify the structure, the control law is designed as follows:

[0063]

[0064] According to the stability analysis, the observer gain and control gain can be defined as l1 = 2ω o , k0 = ω c . The observer bandwidth is ω o , and the controller bandwidth is ω c .

[0065] The active power coordinated control block diagram of the EADRC-based wind turbine is shown in Fig. 1. Figure 4 .

[0066] After the controller is designed, the adjustment of its parameters is crucial to the performance of the controller. In wind power systems, the objective functions of the controller parameter optimization are multiple.

[0067] In another exemplary embodiment of the present application, in the active power control of wind power generation, two optimization objectives are considered. The objective F1 is to minimize the tracking error, which is composed of the integral sum of the time absolute error of the rotor speed and the active power. The objective F2 is to reduce the fatigue load of the wind turbine, which is composed of the sum of the standard deviations of the tower bending moment and the main shaft torque. The total objective F is composed of the weighted sum of F1 and F2. In addition, the constraints on the pitch angle, the generator torque and the decision variables are also considered. The controller parameter optimization model is:

[0068]

[0069] In the formula, b is the parameter matrix of the first-order error active disturbance rejection controller, b 01 and b 02 are the control gains of the two first-order error active disturbance rejection controllers, ω o1 and ω o2 are the bandwidths of the extended state observer of the two first-order error active disturbance rejection controllers, ω c1 and ω c2 are the bandwidths of the two first-order error active disturbance rejection controllers; F is the total objective function; F1 is the control performance sub-objective function; F2 is the fatigue load sub-objective function; h1 is the weighting coefficient of the control performance sub-objective function; h2 is the weighting coefficient of the fatigue load sub-objective function; T0 is the preset initial time; T max is the maximum simulation time; t is the time variable; is the rotor speed reference value; ω r is the rotor speed of the wind turbine; P cmd is the active power reference value; P e is the output power of the wind turbine; std is the standard deviation; M t is the tower bending moment; T s is the main shaft torque; β ref and Δβ ref are the pitch angle command and the pitch angle command change, respectively; T g-ref and ΔT g-ref are the electromagnetic torque command and the electromagnetic torque command change, respectively; s is the time unit, second; Z max and Z minUpper and lower limits of the parameter variable of the first order error active disturbance rejection controller, respectively. is a parameter matrix of the EADRC scheme, represented by the i-th individual in the k-th iteration in the enhanced snow melt optimization algorithm.

[0070] In another exemplary embodiment of the present application, the calculation of the fatigue load of the unit is more complex and has strong nonlinearity. Machine learning represented by bidirectional long short-term memory (BiLSTM) performs well in data prediction and data fitting. Therefore, intelligent modeling of the main shaft torque and the tower bending moment can be realized by using BiLSTM. The present application designs a fatigue load estimation model based on BiLSTM to realize real-time and accurate calculation of the fatigue load change of the wind power system in the optimization process.

[0071] The present application takes the periodic change of the tower bending moment M t and the main shaft torque T s as a typical index of fatigue load.

[0072] In order to facilitate the calculation of the M t and T s of the wind turbine by using the BiLSTM network, the present application constructs and selects 10000 groups of data sets, which include 7 characteristic parameters of pitch angle β, wind speed V, rotor speed ωr, electromagnetic speed ω g , mechanical torque T r , electromagnetic torque T g , and tip speed ratio λ. Although most of the characteristic parameters have a direct impact on the fatigue load, in order to simplify the complexity of the model.

[0073] The present application first performs Pearson correlation analysis on the 7 characteristic parameters and M t and T s , and selects the four characteristic parameters with the highest correlation as the input of the BiLSTM network. The results of the Pearson correlation coefficient are shown in Table 1. From the data in Table 1, β, V, T r and λ are selected as the input characteristic variables of the M t model, and ωr, ω g , T r and T g are selected as the input characteristic variables of the T s model.

[0074] Table 1 Pearson correlation analysis

[0075]

[0076] Long Short-Term Memory (LSTM) is a special type of recurrent neural network designed to solve the problem of gradient vanishing and gradient explosion that traditional recurrent neural networks face when processing long sequence data. Its structure is mainly composed of input gate, forget gate and output gate, as shown in Figure 5 BiLSTM is an enhanced network combining forward LSTM and reverse LSTM. It has independent hidden layers in both forward and reverse directions, and each hidden layer captures features and information in both forward and reverse directions of the data set. The structure of BiLSTM is shown in Figure 6

[0077] In the operation of the wind turbine, we use the BiLSTM network to simulate M t and T s . BiLSTM can accurately capture the complex dependence between each characteristic parameter and M t and T s , so as to quickly estimate the trend of M t and T s . The model equations of M t and T s are as follows:

[0078]

[0079] The above step 103 can be replaced by the following steps 201-206:

[0080] Step 201: Determine the tower bending moment and the main shaft torque as the fatigue load indicators.

[0081] Step 202: Obtain a plurality of operating parameter data of the wind turbine generator.

[0082] Step 203: According to the plurality of operating parameter data, perform Pearson correlation analysis on the plurality of operating parameters and the tower bending moment and the main shaft torque to obtain a Pearson correlation coefficient of each operating parameter with the tower bending moment and a Pearson correlation coefficient of each operating parameter with the main shaft torque.

[0083] Step 204: Select the operating parameters with Pearson correlation coefficients greater than a preset coefficient threshold as the first input characteristic variable corresponding to the tower bending moment and the second input characteristic variable corresponding to the main shaft torque, respectively.

[0084] Step 205: According to the first input characteristic variable value of the wind turbine generator, use the first bidirectional long short-term memory network to obtain the numerical value of the tower bending moment.

[0085] Step 206: According to the second input characteristic variable of the wind turbine generator, use the second bidirectional long short-term memory network to obtain the numerical value of the main shaft torque.

[0086] ​In another exemplary embodiment of the present application, in the power tracking control, the optimization of the controller parameters considers both the tracking performance of the power instruction and the reduction of the fatigue load of the unit, which is a typical multi-objective problem. In order to quickly and accurately locate the optimal controller parameters, an enhanced snow melt optimization algorithm is used to solve the multi-objective function.

[0087] The snow melt optimization algorithm is a new type of bionic optimization algorithm recently introduced, which draws inspiration from nature and solves complex optimization problems by simulating the natural changes of snow. For the multi-objective optimization problem of the active power control system of wind power generation,

[0088] The present application will propose an enhanced snow melt optimization algorithm to obtain the parameters of the EADRC to optimize the overall performance of the active power control of the wind turbine. The enhanced snow melt optimization algorithm includes three improvements, namely Sobol sequence initialization population, golden sine search mechanism and Levy flight strategy. As shown in Figure 7 The detailed process of the enhanced snow melt optimization algorithm is described as follows:

[0089] Step 1: initialize the population size N, the maximum number of iterations K, the search range [Z max , Z min ], two subpopulations N a = N b = N / 2.

[0090] Step 2: Sobol sequence initialization population to obtain a set of initial solutions uniformly distributed between [0, 1].

[0091] Z i (0) = Z min + G n (Z max -Z min )(i = 1, 2, LN) (11)

[0092] In the formula, Z i (0) is the initial position of the i-th individual; G n is the Sobol random sequence distributed in [0, 1].

[0093] Step 3: calculate the fitness value of each individual in the initial population, and record the global optimal solution Z best by comparison. Then build an elite pool through equation (10) to prevent the loss of excellent individuals in the next generation.

[0094]

[0095] In the formula, Z elite (k) is an individual randomly selected from the elite pool; Z b "est represents the global optimum position; Z second (k) and Z third (k) represents the second and third best individuals in the current population; Z c (k) represents the center position of the top 50% individuals with fitness values; is the overall population after rearranging Z i (k) from best to worst; N1is half of the total population N, N1= N / 2.

[0096] Step 4: To balance the exploration and development stages, a dual population mechanism is introduced, where the entire population N is divided into two sub-populations N a and N b .

[0097] Step 5: Enter the exploration stage, Brownian motion is used to simulate the complex process of snow or water turning into vapor. The formula is as follows:

[0098]

[0099] where f(·) is the fitness function. Z is the population individual.

[0100] Step 6: Update the Na and Nb sub-populations.

[0101] N a = N a + 1, N b = N b - 1, if N a < N (15)

[0102] Step 7: Enter the development stage, use the degree-day method to simulate the process of snow melting into water, the focus of this stage is to develop high-quality solutions around the best individual position. The single position update formula is as follows:

[0103]

[0104] where r2 represents a random number between [0, 1]; M(k) represents the rate of snow ablation; BM i (k) is a set of random numbers representing Brownian motion.

[0105] M(k) is described by the degree-day method, the expression is as follows:

[0106]

[0107] where DDF(k) represents the degree-day factor, ranging from 0.35 to 0.6; T(k) represents the average temperature per day.

[0108] Step 8: Update the entire population using the golden sine search mechanism to prevent the population from converging prematurely near the optimal value.

[0109] Z i (k+1) = Z i (k) x sin x1- x2 x sin x1x g1 x Z best -g2 x Z i (k) (i = 1, 2LN; k = 1, 2LK) (18)

[0110] where x1represents a random number between [0, 2π]; x2represents a random number between [0, π]; g1and g2are the golden section coefficients.

[0111] Step 9: After updating the population position, calculate the fitness of the entire population and update the global optimal position Z best . Considering that the original algorithm will widely converge to the vicinity of the optimal value in the later iterations, resulting in a lack of randomness. In this application, the Lévy flight strategy with long and short jump ability is used to mutate the optimal individual Z best to fully increase the diversity of the population and expand the search range. The Lévy escape strategy formula is as follows:

[0112]

[0113] where Z new is the position of the optimal individual after Lévy flight mutation; α is the step length parameter, which is taken as 0.01 here; levy(δ) represents the path following the Lévy distribution.

[0114] Step 10: Compare the fitness values before and after disturbance, and use the greedy strategy to determine whether to update the global optimal value.

[0115]

[0116] where Z b ′ est is the current global optimal solution, Z new is the global optimal solution after mutation, Z best is the global optimal solution before mutation, and f is the fitness function.

[0117] Step 11: Update the elite pool and determine whether the current iteration count has reached the maximum iteration count. If true, Z best is considered the global optimal solution of the entire optimization process. Otherwise, return to step 4.

[0118] Under the solution of the enhanced snow-melt optimization algorithm, the optimal parameters of the EADRC controller can be quickly and accurately obtained under the constraint conditions of minimizing the fatigue load of the unit and the optimal control performance. Reasonable parameter selection can fully exert the performance of the EADRC controller, thereby improving the active power instruction tracking capability of the wind power system.

[0119] In another exemplary embodiment of the present application, the above step 105 of controlling the pitch angle and electromagnetic torque of the wind turbine through two first-order error active disturbance rejection controllers using optimal parameters respectively specifically includes:

[0120] The deviation of the output power of the wind turbine from the active power reference value is input into the first first-order error active disturbance rejection controller, and the pitch angle instruction is output; the optimal parameters of the first first-order error active disturbance rejection controller include b 01 , ω o1 and ω c1 . The deviation of the rotor speed of the wind turbine from the rotor speed reference value is input into the second first-order error active disturbance rejection controller, and the electromagnetic torque instruction is output; the optimal parameters of the second first-order error active disturbance rejection controller include b 02 , ω o2 and ω c2 .

[0121] The overall principle of the method of the present application is shown in Figure 8 The method of the present application is roughly divided into the following three parts:

[0122] 1) The active power coordinated control strategy of the wind turbine based on EADRC is designed to realize stable tracking of the active power instruction.

[0123] 2) The controller parameter optimization problem is realized by using the enhanced snow-melt optimizer to take into account the tracking performance of the active power and the relief of the fatigue load of the unit.

[0124] 3) The establishment of the fatigue estimation model of the wind turbine is realized by using the BiLSTM network.

[0125] Taking a 5MW wind turbine as an example for simulation, the quantitative indexes F, F1 and F2 under three different scenarios of power instruction tracking, pitch angle actuator failure and model mismatch are respectively counted. The statistical results show that the strategy for improving the active power regulation capability of the wind turbine proposed in the present application has excellent power tracking capability in the present example, and the fatigue load is also relieved.

[0126] Based on the same inventive concept, the embodiment of the present application also provides a coordinated optimization control device for improving the active regulation capability of a wind turbine, which is used to implement the coordinated optimization control method for improving the active regulation capability of a wind turbine as described above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme described in the above method, so the specific limitations in one or more coordinated optimization control device embodiments for improving the active regulation capability of a wind turbine provided below can be referred to the limitations of the coordinated optimization control method for improving the active regulation capability of a wind turbine described above, which will not be described here again.

[0127] In one exemplary embodiment, a coordinated optimization control device for improving the active regulation capability of a wind turbine is provided, which includes a controller construction module, an optimization model establishment module, an index calculation module, a parameter optimization module, and a coordinated optimization module.

[0128] The controller construction module is configured to construct a first-order error active disturbance rejection controller; the optimization model establishment module is configured to establish a controller parameter optimization model with multiple objectives of minimizing the fatigue load of the wind turbine and optimizing the control performance; the controller parameter optimization model includes parameters of two first-order error active disturbance rejection controllers; the index calculation module is configured to obtain a fatigue load index value by using a bidirectional long short-term memory network according to the operating parameters of the wind turbine; the parameter optimization module is configured to solve the controller parameter optimization model by using a snow melt optimization algorithm according to the fatigue load index value, the output power and the rotor speed of the wind turbine, and obtain optimal parameters of the two first-order error active disturbance rejection controllers; the snow melt optimization algorithm uses Sobol sequence to initialize a population, uses a golden sine search mechanism to update the entire population, and uses a Levy flight strategy to mutate the optimal individual; and the coordinated optimization module is configured to control the pitch angle and the electromagnetic torque of the wind turbine by using the two first-order error active disturbance rejection controllers with the optimal parameters respectively, so as to adjust the output power and the rotor speed of the wind turbine.

[0129] The technical features of the above embodiments can be combined in any manner. To make the description concise, not all possible combinations of the technical features in the above embodiments are described, but as long as the combinations of the technical features do not exist contradictions, they should be considered as the scope of the present application.

[0130] The principles and implementation modes of the present application are described by using specific examples in this paper, and the above embodiments are only used to help understand the method and its core idea of the present application; at the same time, for those skilled in the art, according to the idea of the present application, the specific implementation mode and application range will be changed. In conclusion, the content of the specification should not be understood as a limitation of the present application.

Claims

1. A coordinated optimal control method for improving the active regulation capability of a wind turbine generator, characterized in that, The application relates to a wind turbine fatigue load control method based on a first-order error active disturbance rejection controller. A first-order error active disturbance rejection controller is constructed; A controller parameter optimization model with the minimum fatigue load and the best control performance of a wind turbine as multiple targets is established; the controller parameter optimization model comprises parameters of two first-order error active disturbance rejection controllers; According to the operating parameters of the wind turbine, a bidirectional long short-term memory network is used to obtain a fatigue load index value; According to the fatigue load index value, the output power and the rotor speed of the wind turbine, a snow melt optimization algorithm is used to solve the controller parameter optimization model, so that the optimal parameters of the two first-order error active disturbance rejection controllers are obtained; the snow melt optimization algorithm uses a Sobol sequence to initialize a population, uses a golden sine search mechanism to update the entire population, and uses a Levy flight strategy to mutate an optimal individual; The output power and the rotor speed of the wind turbine are adjusted by using the two first-order error active disturbance rejection controllers with the optimal parameters to control the pitch angle and the electromagnetic torque of the wind turbine respectively.

2. The coordinated optimal control method for improving the active regulation capability of a wind turbine generator set according to claim 1, characterized in that, The first-order error active disturbance rejection controller comprises an extended state observer and a control law; The extended state observer takes the system error and the control signal output by the control law as inputs, and outputs first and second state variables according to the system error and the control signal output by the control law; when the system error is the deviation of the output power of the wind turbine from the active power reference value, the control signal output by the control law is a pitch angle instruction; when the system error is the deviation of the rotor speed of the wind turbine from the rotor speed reference value, the control signal output by the control law is an electromagnetic torque instruction; The gain of the augmented state observer satisfies l1=2ω o , l1and l2are gains of the augmented state observer, ω o is a bandwidth of the augmented state observer; The control law is according to the first state variable and the second state variable, and is based on an output control signal; wherein, u is the control signal, k0 and b0 are control gains, k0 = ω c , ω c is a bandwidth of the first order error active disturbance rejection controller, z1 is the first state variable, and z2 is the second state variable.

3. The coordinated optimal control method for improving the active regulation capability of a wind turbine generator set according to claim 1, characterized in that, The controller parameter optimization model is as follows: wherein, is a parameter matrix of the first order error active disturbance rejection controller, b 01 and b 02 are control gains of the two first order error active disturbance rejection controllers, ω o1 and ω o2 are extended state observer bandwidths of the two first order error active disturbance rejection controllers, ω c1 and ω c2 are bandwidths of the two first order error active disturbance rejection controllers; F is a total target function; F1 is a control performance sub-target function; F2 is a fatigue load sub-target function; h1 is a weighting coefficient of the control performance sub-target function; h2 is a weighting coefficient of the fatigue load sub-target function; T0 is a preset initial time; T max is the maximum simulation time; t is a time variable; is a rotor speed reference value; ω r is a rotor speed of the wind turbine; P cmd is an active power reference value; P e is a wind turbine output power; std is a standard deviation; M t is a tower bending moment; T s is a main shaft torque; β ref and Δβ ref are a pitch angle command and a pitch angle command change, respectively; T g-ref and ΔT g-ref are the electromagnetic torque command and the electromagnetic torque command change, respectively; s is a unit of time; Z max and Z min are upper and lower limits, respectively, of a parameter variable of the first-order error active disturbance rejection controller.

4. The coordinated optimal control method for improving the active regulation capability of a wind turbine generator set according to claim 1, characterized in that, According to the operating parameters of the wind turbine, a bidirectional long short-term memory network is used to obtain a fatigue load index value, which specifically comprises: The tower moment and the main shaft torque are determined as the fatigue load indexes; A plurality of operating parameter data of the wind turbine are acquired; According to the plurality of operating parameter data, Pearson correlation analysis is performed on the plurality of operating parameters, the tower moment and the main shaft torque, so that a Pearson correlation coefficient of each operating parameter with the tower moment and a Pearson correlation coefficient of each operating parameter with the main shaft torque are obtained; An operating parameter with a Pearson correlation coefficient greater than a preset coefficient threshold value is selected as a first input feature variable corresponding to the tower moment and a second input feature variable corresponding to the main shaft torque; According to the first input feature variable value of the wind turbine, a first bidirectional long short-term memory network is used to obtain the numerical value of the tower moment; According to the second input feature variable of the wind turbine, a second bidirectional long short-term memory network is used to obtain the numerical value of the main shaft torque.

5. The coordinated optimal control method for improving the active regulation capability of a wind turbine generator set according to claim 4, characterized in that, The first input feature variable corresponding to the tower moment comprises a pitch angle, a wind speed, a mechanical torque and a tip speed ratio; The second input feature variable corresponding to the main shaft torque comprises a rotor speed, an electromagnetic speed, a mechanical torque and an electromagnetic torque.

6. The coordinated optimal control method for improving the active regulation capability of a wind turbine generator set according to claim 1, wherein, The specific process of the snow melt optimization algorithm comprises: initializing population parameters; the population parameters comprise an initial population size, a maximum iteration number and a search range; Sobol sequences are used to initialize the initial positions of all individuals in the initial population; According to the initial position of each individual, the fitness value of each individual in the initial population is calculated, and the initial position of the individual with the maximum fitness value is selected as the global optimal solution; An elite pool of the initial population is constructed; A double population mechanism is introduced to divide the entire initial population into two sub-populations; In the exploration stage, Brownian motion is used to simulate the complex process of snow or water turning into vapor, and the two sub-populations are updated; In the development stage, the degree-day method is used to simulate the process of snow melting into water; The position of each individual in the current entire population is updated using the golden sine search mechanism; The fitness value of each individual in the current entire population is calculated, and the global optimal solution is updated; The global optimal solution is mutated using the Levy flight strategy with long-short skip capability to generate a mutated global optimal solution; A greedy strategy is used to determine whether the current global optimal solution is the global optimal solution before mutation or the global optimal solution after mutation; The elite pool is updated; It is determined whether the current iteration count has reached the maximum iteration count, and a comparison result is obtained; If the comparison result is no, the step of introducing the double population mechanism to divide the entire initial population into two sub-populations is returned to; If the comparison result is yes, the current global optimal solution is output.

7. The coordinated optimal control method for improving the active regulation capability of a wind turbine generator set according to claim 6, characterized in that, The elite pool is [Z" best second (k), Z third (k), Z c (k)];​ where Z best is the global optimum solution of the population, Z second (k) and Z third (k) are the second and third best individual positions in the population, Z c (k) is the center position of the top 50% individuals with fitness values, is the overall population after rearranging Z i (k) from best to worst, and N1 is half of the total population size; k = 1, 2…K, and K is the maximum number of iterations.

8. The coordinated optimal control method for improving the active regulation capability of a wind turbine generator set according to claim 6, characterized in that, The greedy strategy is: wherein Z' best is the current global optimum solution, Z new is the global optimum solution after mutation, Z best is the global optimum solution before mutation, and f is the fitness function.

9. The coordinated optimal control method for improving the active regulation capability of a wind turbine generator set according to claim 3, wherein, The pitch angle and electromagnetic torque of the wind turbine are controlled by two first-order error active disturbance rejection controllers using optimal parameters, specifically including: The deviation of the output power of the wind turbine from the active power reference value is input into a first first-order error active disturbance rejection controller, and a pitch angle instruction is output; the optimal parameters of the first first-order error active disturbance rejection controller include b 01 , ω o1 , and ω c1 ; The deviation of the rotor speed of the wind turbine from a rotor speed reference value is input into a second first-order error active disturbance rejection controller, and an electromagnetic torque instruction is output; the optimal parameters of the second first-order error active disturbance rejection controller include b 02 , ω o2 , and ω c2 .

10. A coordinated optimal control device for improving the active regulation capability of a wind turbine generator, characterized in that, Including: A controller construction module is configured to construct a first-order error active disturbance rejection controller; An optimization model establishment module is configured to establish a controller parameter optimization model with the minimum fatigue load of the unit and the best control performance as multiple objectives; the controller parameter optimization model includes parameters of two first-order error active disturbance rejection controllers; An index calculation module is configured to obtain a fatigue load index value by using a bidirectional long short-term memory network according to the operating parameters of the wind turbine; A parameter optimization module is configured to solve the controller parameter optimization model by using a snow melting optimization algorithm according to the fatigue load index value, the output power and the rotor speed of the wind turbine, to obtain optimal parameters of the two first-order error active disturbance rejection controllers; the snow melting optimization algorithm uses Sobol sequence to initialize a population, uses a golden sine search mechanism to update the entire population, and uses a Levy flight strategy to mutate the optimal individual; A coordination optimization module is configured to control the pitch angle and electromagnetic torque of the wind turbine by two first-order error active disturbance rejection controllers using optimal parameters, to adjust the output power and the rotor speed of the wind turbine.

Images