A wind speed prediction variable pitch control system and control method

Abstract

The present invention discloses a wind speed prediction and variable pitch control system and control method. The control method obtains real-time wind farm data and a state detection signal of the wind rotor's pitch angle; performs variational mode decomposition on the real-time wind farm data, then inputs it into a dual-channel convolutional neural network for fusion. The fused features are then input into a bidirectional long-short-term neural network for prediction, resulting in a wind speed prediction result; nonlinearly optimizes the pitch based on the wind speed prediction result, the rotor tip speed ratio, and the wind turbine's speed error to obtain an optimized pitch signal; and controls the wind rotor to change pitch based on the optimized pitch signal and the state detection signal of the pitch angle. By predicting wind speed changes, the frequency and amplitude of pitch change execution are effectively reduced, reducing mechanical fatigue on the wind turbine blades and stabilizing the wind turbine's output.

Description

Technical Field

[0001] The present invention relates to variable pitch control of a wind turbine rotor, and in particular to a wind speed prediction variable pitch control system and a control method. Background Art

[0002] Wind energy is a green, clean, and renewable energy source, attracting widespread attention for its environmentally friendly and renewable characteristics. Wind turbines are the primary means of harvesting wind energy. With the advancement of theoretical research on wind power generation, wind power plants have become widespread worldwide, and the proportion of electricity generated by wind turbines in the global electricity generation is continuously increasing. Due to the unsteady nature of wind, the power output of wind turbines is highly volatile, with a high degree of randomness and intermittency, which undoubtedly increases risks to the safe and stable operation of the power grid. Therefore, incorporating control strategies into wind turbines can reduce the volatility of wind power and ensure the stable operation of the power grid.

[0003] Wind turbine control strategies can be divided into two types based on the rated wind speed: when the wind speed is below the rated wind speed, torque control is used to maximize the power coefficient, that is, the energy captured by the wind turbine; when the wind speed is above the rated wind speed, the wind turbine typically uses variable pitch control, which limits mechanical power by adjusting the angle of the blades, thereby maintaining the output power near the rated power. Variable pitch control is one of the main means of controlling the operation of wind turbines, and is characterized by aerodynamic nonlinearity, frequent switching of operating conditions, and multiple disturbance factors. Wind turbines using traditional variable pitch control systems experience frequent fluctuations in output power due to the pitch adjustment signal changing too quickly with fluctuations in wind speed and output power. This causes the pitch adjustment direction to frequently reverse the direction of wind speed changes. Summary of the Invention

[0004] Purpose of the invention: In view of the above shortcomings, the present invention provides a wind speed prediction pitch control system and control method that can make the output of the wind turbine more stable.

[0005] Technical solution: To solve the above problems, the present invention adopts a wind speed prediction pitch control system, including:

[0006] Anemometer, used to collect real-time data of wind fields;

[0007] A displacement sensor is used to obtain a status detection signal of the pitch angle of the wind rotor;

[0008] The decomposition module is used to pre-process the collected real-time wind field data to obtain wind speed series and meteorological series, and perform variational mode decomposition on the wind speed series and meteorological series to obtain decomposed data;

[0009] The feature fusion module is used to input the decomposed data into the dual-channel convolutional neural network for fusion to obtain the fused features;

[0010] The prediction module is used to input the fused features into the bidirectional long-short-term neural network for prediction to obtain the wind speed prediction result;

[0011] The control signal optimization module is used to perform nonlinear optimization on the pitch according to the wind speed prediction result, the tip speed ratio of the wind rotor and the speed error value of the wind turbine generator to obtain the optimized pitch signal;

[0012] The pitch control module is used to control the wind rotor to change the pitch according to the optimized pitch signal and the state detection signal of the pitch angle.

[0013] Furthermore, the system also includes a tip speed ratio optimizer for calculating the optimal tip speed ratio of the wind rotor at the current wind speed based on the collected real-time wind farm data. It also includes a speed controller for monitoring the rotor speed of the wind turbine in real time and calculating the speed error value of the rotor of the wind turbine.

[0014] Furthermore, the pitch control module obtains an error signal by comparing the obtained state detection signal of the pitch angle with the optimized pitch signal, and then obtains a control signal based on the nonlinear proportional relationship between the pitch angle and the control voltage of the actuator, and controls the wind wheel to perform pitch control according to the control signal.

[0015] Furthermore, the prediction module also includes a model optimization module, which is used to determine the number of training times w, the learning rate ε, and the number of hidden layer neurons o in the bidirectional long-short term neural network by improving the locust optimization algorithm.

[0016] The present invention also adopts a wind speed prediction pitch control method, comprising the following steps:

[0017] (1) Acquire real-time wind farm data and preprocess it to obtain wind speed sequence and meteorological sequence; at the same time, obtain the status detection signal of the wind rotor pitch angle;

[0018] (2) Perform variational mode decomposition on the wind speed series and meteorological series to obtain the decomposed data;

[0019] (3) Input the wind decomposition data into the dual-channel convolutional neural network for fusion to obtain the fused features;

[0020] (4) Input the fused features into the bidirectional long-short term neural network for prediction to obtain the wind speed prediction result;

[0021] (5) Perform nonlinear optimization on the pitch according to the wind speed prediction results, the tip speed ratio of the wind turbine blades, and the speed error of the wind turbine generator to obtain the optimized pitch signal;

[0022] (6) The wind wheel is controlled to change pitch according to the optimized pitch signal and pitch angle status detection signal.

[0023] Furthermore, in step (6), the obtained state detection signal of the pitch angle is compared with the optimized pitch signal to obtain an error signal, and then a control signal is obtained based on the nonlinear proportional relationship between the pitch angle and the control voltage of the actuator, and the wind wheel is controlled to change the pitch according to the control signal.

[0024] Furthermore, the constrained variational mode decomposition formula in step (2) is:

[0025]

[0026]

[0027] Where: S(t) is the undecomposed main signal; u k is the modal function; {ω k}={ω1,ω2,...,ω k} is the center frequency of the obtained K-order mode; δ(t) is the Dirac distribution; * is the convolution; j is the imaginary unit, and t is the time script;

[0028] Introducing the penalty factor α and the Lagrange multiplication operator λ, the constrained variation problem is transformed into an unconstrained variation problem. k 、ω k and λ are iteratively updated to obtain a series of variational mode components.

[0029] Furthermore, the bidirectional long-term short-term neural network in step (4) includes an input layer, a Bi-LSTM layer, a fully connected layer and an output layer; in the input layer, the p fused feature sequences before time t on the dth day are x(tp, d)=(x t-p，d , x t-(p-1)，d ,...,x t-1，d ) as input, q fused feature sequences x(t+q,d)=(x t，d , x t+1，d ,...,x t+q，d ) as output, where p is the time step and q is the prediction step;

[0030] The Bi-LSTM layer performs forward and reverse data feature learning from time 1 to time t, and the two directions are integrated and fed back to the fully connected layer; a Dropout layer is added to prevent overfitting, and the mean square error is selected as the loss function; the number of training times w, the learning rate ε, and the number of hidden layer neurons o are determined;

[0031] The data is subjected to dimensionality reduction processing in the fully connected layer to be converted into one-dimensional data;

[0032] The output layer is responsible for outputting the wind speed prediction result corresponding to the input eigenmode component.

[0033] Furthermore, the improved locust optimization algorithm is used in the Bi-LSTM layer to determine the number of training times w, the learning rate ε, and the number of hidden layer neurons o; specifically, the following steps are included:

[0034] (41) Initialize the number of training times w, learning rate ε, number of hidden layer neurons o, locust race {X i , i=1, 2, 3, ..., n}, maximum number of iterations L;

[0035] The mathematical model of the locust optimization algorithm is:

[0036]

[0037] Among them, N is the number of locusts, c is the decreasing coefficient, ub d is the upper bound of the function s of social power strength in d-dimensional space, lb d is the lower bound of the function s of social power strength in d-dimensional space, x i is the position of the i-th locust, x j is the position of the jth locust, d ij is the distance between the i-th locust and the j-th locust, T d It is the best solution for the locust position in d-dimensional space so far.

[0038] (42) Calculate the fitness of locust individuals, find the non-dominated solution of the current population and update the external archive set, and update the number of iterations l;

[0039] (43) Determine whether the maximum number of iterations has been reached, if so, proceed to step (47), otherwise proceed to step (44);

[0040] (44) Calculate the distance between locust individuals and update the decreasing coefficient;

[0041] (45) Select dominant locust individuals and perform Cauchy mutation on them to make them jump out of the local optimum;

[0042] (46) Determine whether the mutated individual is better than the original individual. If so, update the locust position; otherwise, retain the original solution and return to step (42);

[0043] (47) Output the optimal solution, and output the number of training times w, learning rate ε, and number of hidden layer neurons o in the optimal solution.

[0044] Beneficial Effects: Compared to existing technologies, this invention significantly reduces the frequency and amplitude of pitch control by predicting wind speed variations, minimizing mechanical fatigue on wind turbine blades and stabilizing wind turbine output. The dual-channel convolutional network, compared to traditional single-channel convolutional networks, mitigates data interference, extracts multi-scale information, and achieves more accurate wind speed prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 Shown is a structural principle block diagram of the wind speed prediction pitch control system of the present invention.

[0046] Figure 2 Shown is a schematic diagram of the wind speed prediction system block diagram in the present invention.

[0047] Figure 3 Shown is a flow chart of the improved locust optimization algorithm in the present invention.

[0048] Figure 4 The figure shows a comparison of the front and rear pitch angle changes using the variable pitch control method of the present invention. DETAILED DESCRIPTION

[0049] Example 1

[0050] like Figure 1 As shown, a wind speed prediction and pitch control system in this embodiment includes a wind meter, a wind speed prediction system, a tip speed ratio optimizer, a pitch nonlinear controller, a speed controller, a pitch control system, a pitch actuator, and a displacement sensor. The wind meter transmits the collected wind energy data to the wind speed prediction system and the tip speed ratio optimizer. The tip speed ratio optimizer can calculate the current optimal tip speed ratio, and the wind speed prediction system can predict the wind speed changes in the future.

[0051] like Figure 2 As shown in the figure, the wind speed prediction system performs VMD decomposition on the wind energy data transmitted by the anemometer, then inputs it into a two-channel convolutional neural network (TCNN) for feature fusion extraction, and then inputs it into a bidirectional long short-term memory neural network (Bi-LSTM) for prediction. The Bi-LSTM neural network is optimized using the improved multi-objective locust optimization algorithm (MOGOA) to make the obtained wind speed prediction more accurate.

[0052] The speed controller is used to monitor the rotor speed of the wind turbine in real time, calculate the error value of the rotor speed of the wind turbine, and input it into the pitch nonlinear controller; the pitch nonlinear controller transmits the error signal in the predicted wind speed, tip speed ratio and speed controller into the nonlinear function controller (nonlinear PID controller), generates a pitch signal, and then transmits it to the pitch control system, obtains the state detection signal of the pitch angle through the displacement sensor, and the pitch control system compares the received optimized pitch signal with the state detection signal of the pitch angle to obtain an error signal, obtains a control signal based on the nonlinear proportional relationship between the pitch angle and the control voltage of the actuator, and sends the control signal to the pitch actuator; the pitch actuator is a hydraulic mechanism, and the converted control signal drives the hydraulic actuator to realize the pitch control process.

[0053] Example 2

[0054] A wind speed prediction pitch control method in this embodiment includes the following steps:

[0055] (1) Obtaining real-time wind farm data and preprocessing it to obtain wind speed sequence and meteorological sequence; at the same time, obtaining the status detection signal of the wind rotor pitch angle;

[0056] (2) Perform variational mode decomposition on the preprocessed real-time wind farm data to eliminate random disturbances in the sequence;

[0057] The constrained variational mode decomposition formula is:

[0058]

[0059]

[0060] Where: S(t) is the undecomposed main signal; u k is the modal function; {ω k}={ω1,ω2,...,ω k} is the center frequency of the obtained K-order mode; δ(t) is the Dirac distribution; * is the convolution; j is the imaginary unit, and t is the time script;

[0061] By introducing the penalty factor α and the Lagrange multiplication operator λ, the constrained variational problem is transformed into an unconstrained variational problem, and the augmented expression is obtained as follows:

[0062]

[0063] Initialize the parameters u1, ω1, λ1 and n, the initial value of n is set to 0, set the loop process, let n = n + 1, u k 、ω k Update the value according to the following formula:

[0064]

[0065]

[0066] Where: x(ω) is the Fourier transform of x(t); for u k (ω) The n-th iteration value in the Fourier domain; —ω k The nth iteration value.

[0067] (3) The VMD decomposed data is fed into a two-channel convolutional neural network (TCNN) for fusion, extracting the wind speed sequence and its coupling relationship with the meteorological sequence. By setting the same convolution kernel size to ensure that the two convolutional networks output the same feature length, the two features are concatenated and fused to obtain the fused feature.

[0068] (4) The fused features are input into the long-term time features extracted from the bidirectional long-short term neural network (Bi-LSTM). The forward LSTM analyzes the input variables (including wind speed data and meteorological characteristics) in the forward direction of the time series, and the reverse LSTM analyzes the input variables in the reverse direction of the time series. The outputs of the forward and reverse LSTMs are superimposed and fed back to the fully connected layer. After data dimensionality reduction, the wind speed prediction results are output.

[0069] The bidirectional long short-term neural network consists of an input layer, a Bi-LSTM layer, a fully connected layer, and an output layer:

[0070] In the input layer, the p fused feature sequences before time t on the dth day are

[0071] x(tp,d)=(x t-p，d，xt-(p-1)，d ,...,x t-1，d ) as input, q fused feature sequences x(t+q,d)=(x t，d , x t+1，d ,...,x t+q，d ) as output, where p is the time step and q is the prediction step;

[0072] The Bi-LSTM layer learns data features in both forward and reverse directions from time 1 to time t, integrating the two directions and feeding them back to the fully connected layer. A Dropout layer is added to prevent overfitting, and the mean square error is selected as the loss function. The number of training times w, the learning rate ε, and the number of hidden layer neurons o are determined.

[0073] In the fully connected layer, the data is reduced in dimension and converted into one-dimensional data;

[0074] The output layer is responsible for outputting the wind speed prediction results corresponding to the input eigenmode components.

[0075] In the Bi-LSTM layer, the improved locust optimization algorithm is used to determine the number of training times w, the learning rate ε, and the number of hidden layer neurons o; Figure 3 As shown, the specific steps include:

[0076] (41) Initialize the number of training times w, learning rate ε, number of hidden layer neurons o, locust race {X i , i=1, 2, 3, ..., n}, maximum number of iterations L;

[0077] The mathematical model of the locust optimization algorithm is:

[0078] X i =S i +G i +A i

[0079] Where: X i is the position of the i-th locust; G i is the gravity on the i-th locust; A i is the wind force on the i-th locust; S i is the mutual influence between locust individuals, S i It can be expressed by the following formula:

[0080]

[0081]

[0082] Where: d ij is the distance between the i-th locust and the j-th locust; is the unit vector pointing from the i-th locust to the j-th locust; s is a function representing the strength of social power.

[0083] The improved locust position update model ignores the locust's gravity and assumes that the wind direction always points to the optimal solution, which can be simplified into the following formula:

[0084]

[0085] Among them, N is the number of locusts, c is the decreasing coefficient, ub d is the upper bound of the function s of social power strength in d-dimensional space, lb d is the lower bound of the function s of social power strength in d-dimensional space, x i is the position of the i-th locust, x j is the position of the jth locust, d ij is the distance between the i-th locust and the j-th locust, T d It is the best solution for the locust position in d-dimensional space so far.

[0086] The parameter c can be updated by calculating the following formula.

[0087]

[0088] Where: c max is the maximum value of c; c min is the minimum value of c; l is the current number of iterations; L is the maximum number of iterations.

[0089] (42) Calculate the fitness of locust individuals, find the non-dominated solution of the current population and update the external archive set, and update the number of iterations L;

[0090] (43) Determine whether the maximum number of iterations has been reached, if so, proceed to step (47), otherwise proceed to step (44);

[0091] (44) Calculate the distance between locust individuals and update the decreasing coefficient;

[0092] (45) Select the dominant locust individual and perform Cauchy mutation on it to make it jump out of the local optimum; the formula is:

[0093]

[0094] Where: cauchy is the Cauchy operator; x best is the most abundant locust individual at present; x newbest —The optimal locust individual after using Cauchy mutation.

[0095] (46) Determine whether the mutated individual is better than the original individual. If so, update the locust position; otherwise, retain the original solution and return to step (42);

[0096] (47) Output the optimal solution, and output the number of training times w, learning rate ε, and number of hidden layer neurons o in the optimal solution.

[0097] (5) Perform nonlinear optimization on the pitch according to the wind speed prediction results, the tip speed ratio of the wind turbine blades, and the speed error of the wind turbine generator to obtain the optimized pitch signal;

[0098] (6) The wind wheel is controlled to change pitch according to the optimized pitch signal and pitch angle status detection signal. Figure 4 As shown in the comparison chart of the pitch angles before and after use, the pitch angle changes more frequently and with a larger amplitude before equipment optimization, while the frequency and amplitude of pitch angle changes are significantly smaller after optimization.

Claims

1. A wind speed prediction pitch control method, characterized in that: The following steps are involved: (1) Obtain real-time wind farm data and preprocess it to obtain wind speed sequence and meteorological sequence; at the same time, obtain the status detection signal of the wind rotor pitch angle; (2) Perform variational mode decomposition on the wind speed series and meteorological series to obtain the decomposed data; the constrained variational mode decomposition formula is: ； ； Where: is the undecomposed main signal; is a modal function; }={ } is the center frequency of the obtained K-order mode; is the Dirac distribution; is convolution; is the imaginary unit, For time script; Introducing a penalty factor , Lagrange multiplication operator , transforming the constrained variational problem into an unconstrained variational problem, 、 and Perform iterative updates to obtain a series of variational modal components; (3) Input the decomposed data into a dual-channel convolutional neural network for fusion to obtain the fused features; (4) The fused features are input into the bidirectional long-short-term neural network for prediction to obtain the wind speed prediction result; the bidirectional long-short-term neural network includes an input layer, a Bi-LSTM layer, a fully connected layer and an output layer; in the input layer, the first sky Before the time The fused feature sequence As input, After a moment The fused feature sequence As output, where is the time step, is the prediction step length; The Bi-LSTM layer has a number of layers from 1 to Perform forward and reverse data feature learning at all times, integrate the two directions and feed them back to the fully connected layer; add a Dropout layer to prevent overfitting, and select the mean square error as the loss function; determine the number of training times , learning rate , the number of hidden layer neurons ; The data is subjected to dimensionality reduction processing in the fully connected layer to be converted into one-dimensional data; The output layer is responsible for outputting the wind speed prediction result corresponding to the input eigenmode component; The Bi-LSTM layer uses an improved locust optimization algorithm to determine the number of training times , learning rate , the number of hidden layer neurons ; The specific steps include: (41) Initialization training times , learning rate , the number of hidden layer neurons , locust race , maximum number of iterations ; The mathematical model of the locust optimization algorithm is: ； in, For the number of locusts, is the decreasing coefficient, is a function of the strength of social forces exist The upper bound on the dimensional space, is a function of the strength of social forces exist The lower bound on the dimensional space, For the The location of the locusts, For the The location of the locusts, For the The locust and the The distance between locusts, The locusts are currently located The best solution in dimensional space; (42) Calculate the fitness of locust individuals, find the non-dominated solution of the current population and update the external archive set, and update the number of iterations ; (43) Determine whether the maximum number of iterations has been reached. If so, proceed to step (47); otherwise, proceed to step (44). (44) Calculate the distance between locust individuals and update the decreasing coefficient; (45) Select dominant locust individuals and perform Cauchy mutation on them to make them jump out of the local optimum; (46) Determine whether the mutated individual is better than the original individual. If so, update the locust position; otherwise, retain the original solution and return to step (42). (47) Output the optimal solution and the number of training times in the optimal solution , learning rate , the number of hidden layer neurons ; (5) Based on the wind speed prediction results, the tip speed ratio of the wind turbine blades and the speed error of the wind turbine generator, the pitch is nonlinearly optimized to obtain the optimized pitch signal; (6) The wind rotor is controlled to change pitch according to the optimized pitch signal and pitch angle status detection signal.

2. The wind speed prediction pitch control method according to claim 1, characterized in that: In the step (6), the obtained state detection signal of the pitch angle is compared with the optimized pitch signal to obtain an error signal, and then a control signal is obtained according to the nonlinear proportional relationship between the pitch angle and the control voltage of the actuator, and the wind wheel is controlled to change the pitch according to the control signal.

3. A wind speed prediction pitch control system using the wind speed prediction pitch control method according to claim 1, characterized in that: include: Anemometer, used to collect real-time data of wind fields; A displacement sensor is used to obtain a status detection signal of the pitch angle of the wind rotor; The decomposition module is used to pre-process the collected real-time wind field data to obtain wind speed series and meteorological series, and perform variational mode decomposition on the wind speed series and meteorological series to obtain decomposed data; The feature fusion module is used to input the decomposed data into the dual-channel convolutional neural network for fusion to obtain the fused features; The prediction module is used to input the fused features into the bidirectional long-short-term neural network for prediction to obtain the wind speed prediction result; The control signal optimization module is used to perform nonlinear optimization on the pitch according to the wind speed prediction result, the tip speed ratio of the wind rotor and the speed error value of the wind turbine generator to obtain the optimized pitch signal; The pitch control module is used to control the wind rotor to change the pitch according to the optimized pitch signal and the state detection signal of the pitch angle.

4. The wind speed prediction pitch control system according to claim 3, characterized in that: It also includes a tip speed ratio optimizer for calculating the optimal tip speed ratio of the wind rotor at the current wind speed based on the collected real-time data of the wind field.

5. The wind speed prediction pitch control system according to claim 3, characterized in that: The invention also includes a speed controller for monitoring the rotor speed of the wind generator in real time and calculating a speed error value of the rotor of the wind generator.

6. The wind speed prediction pitch control system according to claim 3, characterized in that: The pitch control module obtains an error signal by comparing the obtained state detection signal of the pitch angle with the optimized pitch signal, and then obtains a control signal based on the nonlinear proportional relationship between the pitch angle and the control voltage of the actuator, and controls the wind wheel to perform pitch control according to the control signal.

7. The wind speed prediction pitch control system according to claim 3, characterized in that: The prediction module also includes a model optimization module, which is used to determine the number of training times in the bidirectional long-term and short-term neural network by improving the locust optimization algorithm. , learning rate , the number of hidden layer neurons .

Images