阅读说明：本技术 一种基于ann自适应滤波的风剪切3倍频纹波抑制方法 (Wind shear 3 frequency multiplication ripple suppression method based on ANN adaptive filtering ) 是由 李志勇 文武斌 王欣 陈有根 于 2021-01-21 设计创作，主要内容包括：本发明公开了一种基于ANN自适应滤波的风剪切3倍频纹波抑制方法,将ANN自适应滤波器嵌入转速双闭环控制系统中,并以外环风轮转速参考值为锁相信号,以转子电流指令信号为学习对象；ANN自适应滤波器中神经元输出和学习对象的误差信号不断调节权值,在满足最小均方差情况下得到内环的转子电流指令信号直流量逼近值,即转子恒电流指令；该指令信号结合PID控制器控制实际转子电流经反馈通道跟踪指令信号,3倍频纹波得到抑制。其应用于风剪切效应下的双馈风机控制系统,由于转子电流与电磁转矩及定子功率间的强耦合关系,在抑制转子电流3倍频纹波的同时能够实现恒电磁转矩控制和定子恒功率控制,避免离心转矩损害机械部件和3倍频电能功率波动涌入电网。(The invention discloses a wind shear 3 frequency multiplication ripple wave suppression method based on ANN adaptive filtering, wherein an ANN adaptive filter is embedded into a rotating speed double closed-loop control system, an outer-loop wind wheel rotating speed reference value is used as a phase-locked signal, and a rotor current instruction signal is used as a learning object; continuously adjusting weights of error signals output by neurons and a learning object in the ANN adaptive filter, and obtaining a direct current value approximation value of a rotor current instruction signal of an inner ring under the condition of meeting the minimum mean square error, namely a rotor constant current instruction; the instruction signal is combined with a PID controller to control the actual rotor current to track the instruction signal through a feedback channel, and 3 frequency multiplication ripples are suppressed. The double-fed fan control system is applied to a double-fed fan control system under the wind shear effect, and due to the strong coupling relation among rotor current, electromagnetic torque and stator power, constant electromagnetic torque control and stator constant power control can be realized while rotor current 3 frequency multiplication ripples are restrained, and the damage of mechanical parts by centrifugal torque and the surge of 3 frequency multiplication electric energy power to a power grid are avoided.)

1. A wind shear 3 frequency multiplication ripple suppression method based on ANN adaptive filtering is characterized by comprising the following steps:

step 1, collecting a wind wheel rotating speed reference value and a rotor current instruction signal in a rotating speed double closed-loop control system of a wind generating set, and respectively using the collected results as a phase-locked signal and a learning object of an ANN adaptive filter;

step 2, continuously adjusting the weight of the neuron by the neuron output of the ANN adaptive filter and an error signal of a learning object;

step 3, the weight of the neuron is adjusted under the condition of meeting the minimum variance, and the adaptive filter outputs a rotor constant current instruction signal of a current inner ring;

step 4, the rotor constant current instruction signal is combined with a PID controller to control the actual rotor current to track the constant current instruction signal through a feedback channel;

2. the ANN adaptive filtering-based wind shear 3 frequency multiplication ripple rejection method according to claim 1, wherein in the step 1, the input signal of the ANN adaptive filter is specifically:

in the step 1.1, the method comprises the following steps of,

setting a wind wheel rotating speed reference value, namely setting the phase locking signal of the ANN filter as omega and the learning object of the ANN adaptive filter as i_rref；

Setting upWhereinIs a component of the ripple that is,is a direct current component;

setting upWhereinIs the ripple component that is learned to be,is the direct current component obtained by learning and is the output of the ANN adaptive filter;

setting a neuron function: (x) x; setting an error signal

Step 1.2, phase-locked frequency multiplication

After omega phase locking and frequency multiplication, sin3 omega t and cos3 omega t signals are obtained through a trigonometric function and are used as reference signals;

in the step 1.3, the step of the method,

setting upWherein w_s、w_c、w₁Is one by oneA neuron weight of the component;

3. the ANN adaptive filtering-based wind shear 3 frequency multiplication ripple rejection method according to claim 1, wherein in the step 2, the specific process of adjusting the neuron weight of the ANN adaptive filter is as follows:

in the step 2.1, the method comprises the following steps of,

setting weight learning rate eta [ eta ]_s η_c η₁]Wherein eta_s、η_c、η₁Respectively adjusting the learning rate in the process of each weight value;

setting damping coefficient xi ═ xi [ ]_s ξ_c ξ₁]In which ξ_s、ξ_c、ξ₁Respectively adjusting the damping coefficient in each weight value process;

setting weight w ═ w_s w_c w₁]；

The orthogonal decomposition component T is set to [ sin3 ω T cos3 ω T1 ]',

using e (t) to adjust the neuron weight value w, the formula is:

w(t+1)＝w(t)+e(t)η*T+ξ(w(t)-w(t-1))

in the step 2.2, the step of the method,

the weight value regulation rule of the neuron is a variance rule which is set as follows:

4. the ANN adaptive filtering-based wind shear 3 frequency multiplication ripple rejection method according to claim 1, wherein in step 3, the minimum mean square error is E [ E (t)²]The dynamic state is close to 0, and the output of the ANN self-adaptive filter is dynamically close to the direct current component of the rotor current instruction;

5. the ANN adaptive filtering-based wind shear 3 frequency multiplication ripple suppression method according to claim 1, wherein in steps 4 and 5, an original inner loop current instruction in a rotating speed double closed loop control system is replaced by a rotor constant current instruction obtained by the ANN to participate in control of inner loop rotor current; the rotor constant current instruction signal is combined with a PID controller to control the actual rotor current to track the instruction signal through a negative feedback channel, and the rotor current ripple is restrained.

Technical Field

The invention relates to a frequency multiplication ripple suppression method, in particular to a ripple suppression method based on ANN adaptive filtering for 3 frequency multiplication ripples caused by wind shear effect.

Background

Accurate real-time detection of ripple is a key technical problem in ripple rejection. At present, a ripple current detection method based on an instantaneous reactive power theory is usually adopted, but the method is complex in calculation, low in precision and poor in instantaneity and self-adaptive capacity. Therefore, it is necessary to find a new ripple detection method by other theories or approaches. The Artificial Neural Network (ANN) has good self-learning and self-adaptive capabilities, and the topological structure and algorithm of the artificial neural network are utilized to provide the ripple current detection method based on the ANN, so that the specific secondary ripple can be dynamically detected accurately in real time, and the ripple suppression effect is ensured.

The main reason for selecting the ANN to realize the self-adaptive filter is that the phase-locked signal passes through a PLL circuit to obtain sin ω t and cos ω t, and then frequency multiplication is carried out to obtain 3 times of sine and cosine signals which are used as reference signals; the net input of the artificial neural network can be formed by linearly combining the net input and the net input, the mapping relation is simple, the structure of the adaptive filter is simple as much as possible in consideration of the easy implementation of the adaptive filter and the real-time requirement of ripple current detection, and a single neuron has a simple structure and certain mapping and adaptive learning capabilities, so that the single neuron is selected as the adaptive filter.

According to the ANN theory, after a plurality of times of iterative learning, the weight value approaches to the optimal value, the mean square error tends to the minimum value at the moment, which means that the ANN self-adaptive filter outputs the current signal which is optimally approximated to be detected and filters other current components under the meaning of the minimum mean square error.

Disclosure of Invention

The invention provides a 3 frequency doubling ripple suppression method based on ANN adaptive filtering, aiming at the problem of wind shearing 3 frequency doubling ripples in a double-fed fan, wherein the application object of the 3 frequency doubling ripple suppression method is a double-fed fan control system under the wind shearing effect, and due to the strong coupling relation among rotor current, electromagnetic torque and stator power, the constant electromagnetic torque control and the stator constant power control can be realized while the rotor current 3 frequency doubling ripples are suppressed, and the damage of mechanical parts by centrifugal torque and the surge of 3 frequency doubling electric energy power to a power grid are avoided.

In order to achieve the above object, the present invention provides a wind shear 3 frequency multiplication ripple suppression method based on ANN adaptive filtering, which includes the following steps:

step 1, collecting a wind wheel rotating speed reference value and a rotor current instruction signal in a rotating speed double closed-loop control system of a wind generating set, and respectively using the collected results as a phase-locked signal and a learning object of an ANN adaptive filter;

step 2, continuously adjusting the weight of the neuron by the neuron output of the ANN self-adaptive filter and an error signal of a learning object;

step 3, obtaining a rotor constant current instruction signal of the current inner ring under the condition of meeting the minimum variance;

step 4, the rotor constant current instruction signal replaces the original rotor current instruction signal and participates in the inner ring rotor current control;

and 5, controlling the actual rotor current to track the command signal through a feedback channel by combining the rotor constant current command signal with a PID (proportion integration differentiation) controller, and inhibiting the rotor current ripple.

Preferably, in step 1, the ANN adaptive filter uses the wind wheel rotation speed reference value as a phase-locked frequency multiplication signal to generate a reference signal; taking the rotor current command signal as a learning object; the specific process of signal input and generation of the ANN adaptive filter is as follows:

in the step 1.1, the method comprises the following steps of,

setting a wind wheel rotating speed reference value, namely setting the phase locking signal of the ANN filter as omega and the learning object of the ANN adaptive filter as i_rrefThe output of the filter is

Setting learning objectWhereinIs a component of the ripple that is,is a direct current component; setting a neuron function: f (x) x, the output isWhereinIs the ripple component that is learned to be,is the direct current component obtained by learning; setting an error signal of

Step 1.2, phase-locked frequency multiplication

After omega phase locking and frequency multiplication, sin3 omega t and cos3 omega t signals, namely reference signals, are obtained through a trigonometric function;

in the step 1.3, the step of the method,

setting upWherein w_s、w_c、w₁The weight of each component neuron;

preferably, in step 2, the output of the ANN adaptive filter is subtracted from the learning object to obtain an error signal, the error signal participates in the learning rule to adjust each weight of the neuron, and the weight is adjusted to be optimal under the condition of minimum mean square error; the specific process of neuron weight adjustment of the ANN adaptive filter is as follows:

in the step 2.1, the method comprises the following steps of,

setting weight learning rate eta [ eta ]_s η_c η₁]Wherein eta_s、η_c、η₁Respectively adjusting the learning rate in the process of each weight value;

setting damping coefficient xi ═ xi [ ]_s ξ_c ξ₁]In which ξ_s、ξ_c、ξ₁Respectively adjusting the damping coefficient in each weight value process;

the orthogonal decomposition component T is set to [ sin3 ω T cos3 ω T1 ]',

setting weight w ═ w_s w_c w₁]；

In the step 2.2, the step of the method,

learning each weight of the neuron adopts a minimum mean square algorithm according to a variance rule, and e (t) is used for adjusting the weight w;

the formula for setting the weight adjustment of the neuron is as follows: w (T +1) ═ w (T) + e (T) η T + ξ (w (T) -w (T-1))

Setting the variance rule for adjusting the weight value as follows:

further preferably, in step 3, the minimum variance condition is E [ E (t)²]The dynamic state is close to 0, the weight value of the neuron is close to the optimal value at the moment, the output of the neuron is close to a learning object, and a rotor constant current instruction signal in the output of the neuronAs the output of the ANN adaptive filter.

Preferably, in step 4 and step 5, the rotor constant current instruction output by the ANN adaptive filter replaces the original inner ring current instruction in the rotating speed double closed-loop control system to participate in the control of the inner ring rotor current; under the control of the PID controller, the actual rotor current tracks the rotor constant current instruction signal to form a closed loop, so that 3 frequency multiplication ripples in the actual rotor current are suppressed.

The invention discloses a wind shear 3 frequency multiplication ripple wave suppression method based on ANN adaptive filtering, wherein an ANN adaptive filter is embedded into a rotating speed double closed-loop control system, a rotating speed reference value of a wind wheel in a rotating speed outer loop is taken as a phase-locked signal, and a rotor current instruction signal is taken as a learning object; continuously adjusting the weight of the neuron by the output of the neuron in the ANN adaptive filter and an error signal of a learning object, and obtaining a direct current value approximation value of a rotor current instruction signal of a current inner ring under the condition of meeting the minimum mean square error, namely a rotor constant current instruction of the current inner ring; the rotor constant current instruction signal is combined with a PID controller to control the actual rotor current to track the instruction signal through a feedback channel, and 3 frequency multiplication ripples are suppressed. The application object of the double-fed fan control system is a double-fed fan control system under the wind shear effect, and due to the strong coupling relation among rotor current, electromagnetic torque and stator power, constant electromagnetic torque control and stator constant power control can be realized while rotor current 3-frequency-multiplication ripples are restrained, and the damage of mechanical parts by centrifugal torque and the surge of 3-frequency-multiplication electric energy power to a power grid are avoided.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the structures shown in the drawings without creative efforts.

Fig. 1 is a schematic flow chart of a 3-fold frequency ripple suppression method based on an ANN adaptive filter according to an embodiment of the present invention;

FIG. 2 is a schematic diagram of an ANN adaptive filter according to an embodiment of the present invention

FIG. 3 is a block diagram of a rotational speed dual closed-loop control system using the method of the present invention in the embodiment of the present invention

The objects, features and advantages of the present invention will be further explained with reference to the accompanying drawings.

Detailed Description

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

In addition, the technical solutions in the embodiments of the present invention may be combined with each other, but it must be based on the realization of those skilled in the art, and when the technical solutions are contradictory or cannot be realized, such a combination of technical solutions should not be considered to exist, and is not within the protection scope of the present invention.

As shown in fig. 1 to 3, a wind shear 3 frequency multiplication ripple suppression method based on ANN adaptive filtering includes the following steps:

step 1, collecting a wind wheel rotating speed reference value and a rotor current instruction signal in a rotating speed double closed-loop control system of a wind generating set, and respectively using the collected results as a phase-locked signal and a learning object of an ANN adaptive filter;

step 2, continuously adjusting the weight of the neuron by the neuron output of the ANN adaptive filter and an error signal of a learning object;

step 3, outputting a rotor constant current instruction signal of the current inner ring by the self-adaptive filter under the condition of meeting the minimum variance;

step 4, the rotor constant current instruction signal replaces the original rotor current instruction signal and participates in the inner ring rotor current control;

and 5, controlling the actual rotor current to track the command signal through a negative feedback channel by combining the rotor constant current command signal with the PID controller, and inhibiting the rotor current ripple.

Step 1, collecting a wind wheel rotating speed reference value and a rotor current instruction signal in a rotating speed double closed-loop control system of the wind generating set, and respectively using the collected results as a phase-locked signal and a learning object of an ANN adaptive filter; the reference value of the rotating speed of the wind wheel can be obtained by an MPPT control system of the wind generating set, and additional sensor equipment is not needed; a rotor current instruction signal in the rotating speed double closed-loop control system needs to be acquired by adding a sensor; setting a wind wheel rotating speed reference value, namely setting an ANN filter phase-locked signal as omega and a rotor current command signal as an ANN adaptive filter learning object i_rrefBecause the rotor current command signal contains 3 frequency multiplication ripples, so that it can make the rotor current command signal possess high speed and high speedWhereinIs a component of the ripple that is,is a direct current component;

the wind wheel rotating speed reference value omega is used as a phase-locked signal for phase-locking frequency multiplication, and then signals of sin3 omega and cos3 omega are obtained through a trigonometric function and are used as reference signals of the ANN self-adaptive filter;

order toAs input to a neuron, where w_s、w_c、w₁Is the neuron weight of each component.

Setting the neuron function f (x) x, the output of the neuronTherefore, the neuron outputs after learningAlso contains 3 frequency multiplication ripple components, such thatWhereinIs the ripple component that is learned to be,is the direct current component obtained by learning and is the output of the ANN adaptive filter; at this time, the error signal may be expressed as

In step 2, the adjustment of the neuron weight requires the participation of an error signal e (t), and also requires the setting of a learning rate, so that the learning rate η of the weight is [ η ]_s η_c η₁]Wherein eta_s、η_c、η₁The learning rates in the process of adjusting the weights are respectively, and the value interval of each learning rate is (0, 1)]Too high a learning rate may lead to a non-convergence of the conditioning process, learningToo low a learning rate can result in too slow a learning process; in addition, a damping coefficient term needs to be set to dynamically increase learning efficiency, and a damping coefficient xi ═ xi is set_s ξ_c ξ₁]In which ξ_s、ξ_c、ξ₁The damping coefficients in the process of adjusting each weight are respectively, and the value interval of each damping coefficient is (0, 1)](ii) a Let weight w be [ w ]_s w_cw₁]The orthogonal decomposition component T ═ sin3 ω T cos3 ω T1]' then, the formula for adjusting or learning each weight value of the sine and cosine signal neurons is: w (T +1) ═ w (T) + e (T) η T + ξ (w (T) -w (T-1)); the weight of the neuron is regulated according to the ruleThe absolute value of the error e (t) is ensured to be continuously reduced in the learning process.

In step 3, the minimum mean square error is E [ E (t)²]Approach 0, when the output of the neuron isApproaches to learning object i_rrefI.e. the weight adjustments are optimal, the output of the ANN adaptive filterAlso approximating ripple component in learning objectThis means that, as shown in fig. 2, the learning object input by the ANN adaptive filter contains a signal with a dc component and a ripple component, and the output is a signal with only a dc component, so as to complete the filtering, i.e., the purpose of filtering out the ripple component.

In steps 4 and 5, as shown in FIG. 3, the primary rotor current command signal i_rrefFiltering out 3 frequency multiplication ripple waves by adaptive filter to generate new current instruction, i.e. rotor constant current instruction signalIncorporate PID controllers to participate in rotating the inner ringThe control of the sub-current is realized by that the actual rotor current continuously tracks the constant current command signal of the rotor through a negative feedback channelTherefore, the actual rotor current only contains direct current, and the suppression of the rotor current 3 frequency multiplication ripple is completed. Due to the strong coupling relation among the rotor current, the electromagnetic torque and the stator power, the constant electromagnetic torque control and the stator constant power control can be realized while the rotor current 3 frequency multiplication ripple wave is restrained, and the damage of mechanical parts by centrifugal torque and the surge of 3 frequency multiplication electric energy power to the power grid are avoided.

The above description is only a preferred embodiment of the present invention, and is not intended to limit the scope of the present invention, and all modifications and equivalents of the present invention, which are made by the contents of the present specification and the accompanying drawings, or directly/indirectly applied to other related technical fields, are included in the scope of the present invention.