阅读说明：本技术 基于自适应神经网络有限时间控制的垂直轴风电机组悬浮控制方法 (Suspension control method of vertical axis wind turbine generator based on self-adaptive neural network finite time control ) 是由 邱雅兰 蔡彬 谌义喜 田玉蓉 褚晓广 于 2021-08-21 设计创作，主要内容包括：本发明涉及基于自适应神经网络有限时间控制的垂直轴风电机组悬浮控制方法,属风力发电领域。风速达到切入风速时,悬浮控制器采用全局快速终端滑模控制和连续有限时间控制策略控制悬浮电流,使垂直轴风电机组旋转体上升至并保持在悬浮平衡点；当实现稳定悬浮后,为应对风速风向随机性造成的不确定未知时变干扰,悬浮控制器改用自适应神经网络全局快速终端滑模控制策略,实时在线估计系统不确定未知项,同时仍采用连续有限时间控制策略控制悬浮电流,使旋转体在机组发电过程中保持稳定悬浮在平衡点。本发明消除了传统终端滑模控制和有限时间控制的抖振问题,具有跟踪快速、抗干扰能力强、运行稳定等优点,尤其适合于低风速磁悬浮垂直轴风电机组。(The invention relates to a suspension control method of a vertical axis wind turbine based on self-adaptive neural network finite time control, belonging to the field of wind power generation. When the wind speed reaches the cut-in wind speed, the suspension controller controls the suspension current by adopting a global fast terminal sliding mode control and continuous finite time control strategy, so that the vertical axis wind turbine generator rotator rises to and is kept at a suspension balance point; after stable suspension is realized, in order to deal with uncertain unknown time-varying interference caused by randomness of wind speed and wind direction, a suspension controller uses a self-adaptive neural network global fast terminal sliding mode control strategy instead, real-time online estimation of uncertain items of a system is carried out, and meanwhile, a continuous finite time control strategy is still used for controlling suspension current, so that the rotating body is kept stably suspended at a balance point in the generating process of the unit. The method eliminates the buffeting problem of the traditional terminal sliding mode control and the limited time control, has the advantages of high tracking speed, strong anti-interference capability, stable operation and the like, and is particularly suitable for the low-wind-speed magnetic suspension vertical axis wind turbine generator.)

1. The suspension control method of the vertical axis wind turbine generator based on the self-adaptive neural network finite time control comprises the steps that the vertical axis wind turbine generator comprises a magnetic suspension vertical axis wind turbine, a suspension control system, an air gap sensor, a wind wheel and a rotating shaft; the magnetic suspension vertical shaft wind driven generator comprises a permanent magnet direct drive type wind driven generator and a magnetic suspension disc type motor; the permanent magnet direct-drive wind driven generator comprises a stator and a rotor; the magnetic suspension disc type motor comprises a disc stator and a disc rotor; the disc stator consists of a disc stator iron core and a suspension winding, and the suspension winding is a direct-current excitation winding; the suspension control system consists of a suspension converter and a suspension controller thereof, the suspension converter is connected with the suspension winding, and the suspension controller comprises an outer ring suspension air gap tracking controller and an inner ring suspension current tracking controller; the rotor of the permanent-magnet direct-drive wind driven generator, the disc rotor of the magnetic suspension disc type motor, the wind wheel and the rotating shaft are collectively called as a rotating body; the method is characterized by comprising the following steps:

step 1, when the wind speed reaches the cut-in wind speed of the magnetic suspension vertical axis wind turbine, an outer ring suspension air gap tracking controller of a suspension controller adopts a global fast terminal sliding mode control strategy with a finite time convergence characteristic, and an inner ring suspension current tracking controller of the suspension controller adopts a continuous finite time control strategy to control the current of a suspension winding, so that a rotating body is suspended upwards to and kept at a suspension balance point according to a preset reference track to realize stable suspension; the specific method comprises the following steps:

11) designing the outer ring suspension air gap tracking controller by adopting a global fast terminal sliding mode control strategy:

A1. designing a global fast terminal sliding mode surface as follows:

in the formula, e₁For floating air gap tracking error: e.g. of the type₁＝δ_ref-δ，δ_refIs a suspended air gap reference value, and delta is a suspended air gap measured value; alpha is alpha₀、β₀>0，p₀And q is₀Is a positive odd number, and p₀>q₀；

The derivation for equation (1) is:

in the formula (I), the compound is shown in the specification,andare respectively delta and delta_refThe second derivative with respect to time t;

according to the upward suspension suction force applied to the rotating body in the axial direction, the downward self-gravity of the rotating body and the external interference force, the kinetic equation of the rotating body in the vertical direction can be obtained as follows:

wherein m is the mass of the rotating body, and g is the gravitational acceleration; f. of_d(t) is unknown time-varying interference; k is mu₀N²S/4, wherein₀The magnetic pole surface of the disc type stator core is in vacuum magnetic conductivity, S is the effective area of the magnetic pole surface of the disc type stator core, and N is the number of turns of the suspension winding; i is_fThe current of the suspension winding is called suspension current;

from formula (3):

wherein d (t) ═ f_d(t)/m，f(x)＝-k/(mδ²)，d (t), f (x), u (t) respectively represent the system uncertain item, the system known item and the output of the terminal sliding mode controller;

when formula (4) is substituted for formula (2), it is possible to obtain:

A2. and (3) obtaining the output of the global fast terminal sliding mode controller:

taking a terminal sliding mode index approach law as follows:

in the formula (I), the compound is shown in the specification,and η are both positive real numbers, p and q are both positive odd numbers, and p>q；

Comparing formula (5) with formula (6), and using u_TSMCAnd replacing u (t), the output of the global fast terminal sliding mode controller can be obtained as follows:

A3. the output u of the global fast terminal sliding mode controller obtained in the step A2 in the formula (7)_TSMCAfter the absolute value of the current is calculated, the output of the outer ring suspension air gap tracking controller is obtained and is made to be the reference value I of the suspension current_ref：

12) An inner ring suspension current tracking controller is designed by adopting a continuous finite time control strategy:

the reference value I of the levitation current_refMinus the actual value I thereof_fTo obtain its error e₂(t), namely: e.g. of the type₂(t)＝I_ref-I_f(ii) a According to the finite time control theory, the reference value of the voltage U of the levitation winding is designed as follows:

in the formula, λ₁∈(0,1)，k₁>0，k₂>0，λ₁、k₁、k₂Are all controller adjustable parameters; tanh (·) represents a hyperbolic tangent function;

a continuous finite time controller formed by the formula (9) is used as the inner ring suspension current tracking controller;

13) tracking the output u of the inner loop suspension current tracking controller_C-FTCSending the current to a PWM module to generate a driving signal of the suspension converter so as to control the suspension current I_fSuspending the rotating body upwards to ensure that the rotating body is stabilized at an equilibrium point;

step 2, after stable suspension is realized, an outer ring suspension air gap tracking controller of the suspension controller uses a self-adaptive neural network global fast terminal sliding mode control strategy instead, and an inner ring suspension current tracking controller of the suspension controller still uses a continuous finite time control strategy to control the suspension current so as to ensure that the rotating body keeps stable suspension at a balance point; the specific method comprises the following steps:

21) and (3) estimating d (t) by using a system uncertainty unknown term d (t) in the radial basis function neural network approximation formula (7):

B1. determining the structure of the radial basis function neural network:

the radial basis function neural network consists of 1 input layer and 1 radial basis function neural networkThe hidden layer and the 1 output layer are formed, the input layer comprises 2 neurons, and corresponding input vectors are as follows:wherein the content of the first and second substances,is e₁The derivative with respect to time t; the hidden layer comprises n neurons, and the output layer comprises 1 neuron;

B2. selecting a Gaussian function as an activation function of the hidden layer, and outputting the hidden layer as follows:

in the formula, h_jOutput of the jth neuron of the hidden layer, c_j＝[c_j1,c_j2]^TIs the central vector of the j-th neuron Gaussian basis function of the hidden layer, | | E-c_jI is the Euclidean norm measuring the input vector E and the j-th neuron central vector of the hidden layer, b_jIs the width vector of the gaussian basis function of the jth neuron of the hidden layer;

B3. calculating the output of the output layer by the weighted sum of the output value of the hidden layer and the weight from the hidden layer to the output layer, and enabling the output of the output layer to be the estimated value of the system uncertain unknown item d (t) in the formula (7)Then there are:

in the formula (I), the compound is shown in the specification,a weight vector representing the output layer,h＝[h₁,h₂,…,h_n]^Tan output vector representing the hidden layer, wherein h_j(j ═ 1,2 …, n) determined by formula (10);

22) obtaining the output of the global fast terminal sliding mode controller of the adaptive radial basis function neural network:

using the output of the radial basis function neural network according to equation (7)Replacing the uncertain items d (t) in the global fast terminal sliding mode controller and using u_ARBF-TSMCSubstitution u_TSMCThe output of the global fast terminal sliding mode controller of the self-adaptive radial basis function neural network can be obtained as follows:

in the formula (I), the compound is shown in the specification,obtained from the formula (11);

23) determining the self-adaptive law of updating the weight of the radial basis function neural network as follows:

wherein γ > 0;

24) the output u of the global fast terminal sliding mode controller of the adaptive radial basis function neural network obtained in the step 22) in the formula (12)_ARBF-TSMCThe absolute value of the outer ring levitation air gap tracking controller is obtained, and the output of the outer ring levitation air gap tracking controller is made to be the reference value I of the levitation current_Aref：

25) Suspending the suspensionReference value I of the floating current_ArefMinus the actual value I thereof_fTo obtain its error e₃(t), namely: e.g. of the type₃(t)＝I_Aref-I_f(ii) a According to the finite time control theory, the reference value of the voltage U of the levitation winding is designed as follows:

in the formula, λ₂∈(0,1)，k₃>0，k₄>0，λ₂、k₃、k₄Are all controller adjustable parameters; tanh (·) represents a hyperbolic tangent function;

the formula (22) forms a continuous finite time controller which is used as the inner ring suspension current tracking controller;

26) tracking the output u of the inner loop suspension current tracking controller_AC-FTCSending the current to a PWM module to generate a driving signal of the suspension converter so as to control the suspension current I_fAnd keeping the rotating body stably suspended at the balance point.

2. The adaptive neural network finite time control-based vertical axis wind turbine generator suspension control method according to claim 1, wherein the specific method for determining the adaptive law of the radial basis neural network weight update in step 23) is as follows:

C1. let the system uncertain element d (t) be represented as:

d(t)＝w^*Th+ε (13)

in the formula, w^*The ideal weight of the output layer of the radial basis function neural network is obtained; epsilon is the approximation error of the radial basis function neural network, and the error can be limited to be small enough based on the infinite precision approximation function of the radial basis function neural network, and epsilon is more than or equal to | epsilon |_Nε N is the minimum supremum of ε, a bounded positive real number;

according to equations (11) and (13), the approximation error of the system uncertainty unknown term d (t) can be expressed as:

in the formula (I), the compound is shown in the specification,the deviation of the network weight value is obtained; under the action of the radial basis function neural network, the bounded real number sigma is not less than 0, so that the approximation error meets the requirement

By u_ARBF-TSMCIn place of u (t), formula (12) and formula (13) are substituted for formula (5), and the corresponding sliding mode approach law is obtained by combining formula (14):

C2. constructing a Lyapunov function:

wherein γ > 0;

derivation of equation (16) and substitution of equation (15) gives:

C3. taking the adaptive law of updating the weight of the radial basis function neural network as follows:

by substituting formula (18) for formula (17), there are

Let η be σ/| s^q/pZeta + zeta >0, then

Since q and p are both positive odd numbers, in formula (20)If true; approximation error satisfaction of uncertain unknown items of the system due to radial basis function neural networkAnd epsilon can be limited to be small enough, and the design of a neural network and the selection of controller parameters can ensure that sigma s plus epsilon s in the formula is more than or equal to 0, namely the requirement ofThe system is thus stable.

Technical Field

The invention relates to a control method, in particular to a suspension control method of a vertical axis wind turbine generator based on self-adaptive neural network finite time control, and belongs to the technical field of wind power generation.

Background

At present, a high-power wind driven generator takes a horizontal shaft wind driven generator as a mainstream product. However, the horizontal axis wind turbine has inherent defects of wind-to-yaw requirement, large starting resisting moment, complex and difficult control, inconvenient installation, high cost and the like, influences the healthy development of the horizontal axis wind turbine, and is particularly difficult to meet the low wind speed starting requirement of a weak wind type wind power plant.

The magnetic suspension vertical axis wind driven generator has the advantages of low starting wind speed, simplicity and convenience in installation, no need of a yaw device and the like, can be used for a wind power plant with low wind speed and frequent wind direction change (no need of wind alignment due to the vertical axis wind driven generator), and is a key direction for future wind power development.

However, at present, the magnetic levitation technology still faces many challenges, such as the magnetic levitation system has the characteristics of open loop instability, high nonlinearity, strong coupling and the like. Particularly, considering that the actual working conditions of the wind turbine generator are variable, uncertainty and external interference always exist in the system, and an accurate mathematical model is difficult to establish, which brings huge challenges to the stable operation and effective control of the magnetic suspension system. The model linearization based on the balance point is the earliest suspension control strategy, but the linearization method can only be converged in a small range, and finally the tracking performance of the system is rapidly deteriorated along with the deviation from the balance point, and the robustness of the system is poor; some nonlinear control strategies, such as adaptive control, sliding mode control, robust control and the like, can improve the robustness of the system, but the control strategies are based on progressive stability analysis, and progressive stability means that the system is converged within an infinite time and can never be converged to a balance point, and the closer to the balance point, the slower the convergence speed of the system is. Moreover, in practical engineering applications, it is often difficult to achieve satisfactory results with progressively stable convergence rates. By finite time stabilization, it is meant that the system can converge to the point of equilibrium within a finite time, as opposed to progressive stabilization. In recent years, with the proposition and the perfection of a finite time homogeneous theory and a finite time Lyapunov stability theory, some finite time controllers are more clear and concise in form and widely applied. The terminal sliding mode control is used as an important branch of a limited time control strategy, and a specially designed nonlinear switching surface is adopted, so that the tracking error of the system can reach a balance point in limited time along a sliding mode. However, no matter terminal sliding mode control or traditional finite time control, the buffeting problem is brought to the system by the essential discontinuous characteristic of the symbol function item in the controller.

The self-adaptive neural network finite time control realizes the finite time convergence of the system through the action of a finite time controller, and realizes the estimation of uncertain unknown items of the system through the real-time approximation action of a self-adaptive neural network model. Compared with the conventional control strategy, the finite time control has strong robustness and high convergence speed, can realize the finite time convergence of the system, can compensate the self-adaptive neural network on line in time when unknown time-varying interference exists in the system, reduces deviation, and is particularly suitable for the control of a magnetic suspension wind power system with nonlinearity, strong coupling and unknown time-varying interference (wind speed, wind direction volatility and uncertainty). However, currently, the application research of the self-adaptive neural network limited time control in the aspect of magnetic suspension wind turbine generators is very little.

Disclosure of Invention

The main purposes of the invention are as follows: aiming at the defects and blanks of the prior art, the invention provides a suspension control method of a magnetic suspension vertical axis wind turbine, which ensures the stable start and reliable operation of the magnetic suspension vertical axis wind turbine under the conditions that the magnetic suspension vertical axis wind turbine is subjected to unknown time-varying interference and system parameters are uncertain by adopting a finite time control strategy and utilizing the real-time online compensation function of an adaptive neural network.

In order to achieve the above object, the magnetic suspension vertical axis wind turbine generator set of the present invention includes: the magnetic suspension vertical axis wind driven generator comprises a magnetic suspension vertical axis wind driven generator, a suspension control system, an air gap sensor, a wind wheel, a rotating shaft, a supporting frame and the like; the magnetic suspension vertical shaft wind driven generator comprises a permanent magnet direct drive type wind driven generator and a magnetic suspension disc type motor.

The permanent magnet direct-drive wind driven generator comprises a stator and a rotor;

the magnetic suspension disc type motor is positioned below the permanent magnet direct-drive type wind driven generator and comprises a disc stator and a disc rotor; the disc stator is composed of a disc stator iron core and a suspension winding, and the suspension winding is a direct-current excitation winding.

The suspension control system consists of a suspension converter and a suspension controller, and the suspension converter is connected with the disc type suspension winding; the suspension controller comprises an outer ring suspension air gap tracking controller and an inner ring suspension current tracking controller.

The rotor of the permanent-magnet direct-drive wind driven generator, the disc rotor of the magnetic suspension disc type motor, the wind wheel and the rotating shaft are collectively called as a rotating body.

The invention relates to a suspension control method of a vertical axis wind turbine generator based on adaptive neural network finite time control, which comprises the following steps:

step 1, when the wind speed reaches the cut-in wind speed of the magnetic suspension vertical axis wind turbine, an outer ring suspension air gap tracking controller of the suspension controller adopts a global fast terminal sliding mode control strategy with a finite time convergence characteristic, and an inner ring suspension current tracking controller of the suspension controller adopts a continuous finite time control strategy to control the current of the disc type suspension winding, so that the rotating body is suspended upwards to and kept at a suspension balance point according to a preset reference track to realize stable suspension. The specific method comprises the following steps:

11) designing the outer ring suspension air gap tracking controller:

A1. designing a global fast terminal sliding mode surface as follows:

in the formula, e₁For floating air gap tracking error: e.g. of the type₁＝δ_ref-δ，δ_refIs a suspended air gap reference value, and delta is a suspended air gap measured value; alpha is alpha₀、β₀>0，p₀And q is₀Is a positive odd number, and p₀>q₀。

The derivation for equation (1) is:

in the formula (I), the compound is shown in the specification,andare respectively delta and delta_refSecond derivative with respect to time t.

According to the upward suspension suction force applied to the rotating body in the axial direction, the downward self-gravity of the rotating body and the external interference force, the kinetic equation of the rotating body in the vertical direction can be obtained as follows:

wherein m is the mass of the rotating body, and g is the gravitational acceleration; f. of_d(t) is unknown time-varying interference; k is mu₀N²S/4, wherein₀The magnetic pole surface effective area is the vacuum magnetic conductivity, S is the magnetic pole surface effective area of the disc type stator iron core, and N is the number of turns of the suspension winding; i is_fThe current of the levitation winding is called levitation current.

From formula (3):

wherein d (t) ═ f_d(t)/m，f(x)＝-k/(mδ²)，d (t), f (x), u (t) respectively represent the system uncertain item, the system known item and the output of the terminal sliding mode controller;

when formula (4) is substituted for formula (2), it is possible to obtain:

A2. and (3) obtaining the output of the global fast terminal sliding mode controller:

taking a terminal sliding mode index approach law as follows:

in the formula (I), the compound is shown in the specification,and η is a positive real number, p and q are positive odd numbers, and p>q。

Comparing formula (5) with formula (6), and using u_TSMCAnd replacing u (t), the output of the global fast terminal sliding mode controller can be obtained as follows:

A3. the output u of the global fast terminal sliding mode controller obtained in the step A2 in the formula (7)_TSMCAfter the absolute value of the current is calculated, the output of the outer ring suspension air gap tracking controller is obtained and is made to be the reference value I of the suspension current_ref：

12) An inner ring suspension current tracking controller is designed by adopting a continuous finite time control strategy:

the reference value I of the levitation current_refMinus the actual value I thereof_fTo obtain its error e₂(t), namely: e.g. of the type₂(t)＝I_ref-I_f(ii) a According to the finite time control theory, the reference value of the voltage U of the levitation winding is designed as follows:

in the formula, λ₁∈(0，1)，k₁>0，k₂>0，λ₁、k₁、k₂Are all controller adjustable parameters; tanh (·) represents a hyperbolic tangent function;

the formula (9) constitutes a continuous finite time controller as the inner loop levitation current tracking controller.

13) Tracking the output u of the inner loop suspension current tracking controller_C-FTCSending the current to a PWM module to generate a driving signal of the suspension converter so as to control the suspension current I_fThe rotating body is suspended upwards to ensure that the rotating body is stabilized at an equilibrium point.

Step 2, after stable suspension is realized, an outer ring suspension air gap tracking controller of the suspension controller uses a self-adaptive neural network global fast terminal sliding mode control strategy instead, and an inner ring suspension current tracking controller of the suspension controller still uses a continuous finite time control strategy to control the suspension current so as to ensure that the rotating body keeps stable suspension at a balance point; the specific method comprises the following steps:

21) and (3) estimating d (t) by using a system uncertainty unknown term d (t) in the radial basis function neural network approximation formula (7):

B1. determining the structure of the radial basis function neural network:

the radial basis function neural network is composed of 1 input layer, 1 hidden layer and 1 output layer, wherein the input layer comprises 2 neurons, and the corresponding input vectors are as follows:wherein the content of the first and second substances,is e₁The derivative with respect to time t; the hidden layer includes n neurons, and the output layer has 1 neuron.

B2. Selecting a Gaussian function as an activation function of the hidden layer, and outputting the hidden layer as follows:

in the formula, h_jOutput of the jth neuron of the hidden layer, c_j＝[c_j1,c_j2]^TIs the central vector of the j-th neuron Gaussian basis function of the hidden layer, | | E-c_jI is the Euclidean norm measuring the input vector E and the j-th neuron central vector of the hidden layer, b_jIs the width vector of the gaussian basis function of the jth neuron of the hidden layer.

B3. Calculating the output of the output layer by the weighted sum of the output value of the hidden layer and the weight from the hidden layer to the output layer, and enabling the output of the output layer to be the estimated value of the system uncertain unknown item d (t) in the formula (7)Then there are:

in the formula (I), the compound is shown in the specification,a weight vector representing the output layer, h ═ h₁,h₂,…，h_n]^TAn output vector representing the hidden layer, wherein h_j(j ═ 1,2 …, n) was determined by formula (10).

22) Obtaining the output of the global fast terminal sliding mode controller of the adaptive radial basis function neural network:

using the output of the radial basis function neural network according to equation (7)Replacing the uncertain items d (t) in the global fast terminal sliding mode controller and using u_ARBF-TSMCSubstitution u_TSMCAvailable adaptive radial basis function neural networkThe output of the network global fast terminal sliding mode controller is as follows:

in the formula (I), the compound is shown in the specification,obtained from the formula (11).

23) Determining the self-adaptive law of updating the weight of the radial basis function neural network:

C1. let the system uncertain element d (t) be represented as:

d(t)＝w^*Th+ε (13)

in the formula, w^*The ideal weight of the output layer of the radial basis function neural network is obtained; epsilon is the approximation error of the radial basis function neural network, and the error can be limited to be small enough based on the infinite precision approximation function of the radial basis function neural network, and epsilon is more than or equal to | epsilon |_N，ε_NIs the smallest supremum of epsilon and is a bounded, positive real number.

According to equations (11) and (13), the approximation error of the system uncertainty unknown term d (t) can be expressed as:

in the formula (I), the compound is shown in the specification,the deviation of the network weight value is obtained; under the action of the radial basis function neural network, the bounded real number sigma is not less than 0, so that the approximation error meets the requirement

By u_ARBF-TSMCIn place of u (t), formula (12) and formula (13) are substituted for formula (5), and the corresponding sliding mode approach law is obtained by combining formula (14):

C2. constructing a Lyapunov function:

wherein γ > 0.

Derivation of equation (16) and substitution of equation (15) gives:

C3. taking the adaptive law of updating the weight of the radial basis function neural network as follows:

by substituting formula (18) for formula (17), there are

Let η be σ/| s^q/pZeta + zeta >0, then

Since q and p are both positive odd numbers, in formula (20)If true; approximation error satisfaction of uncertain unknown items of the system due to radial basis function neural networkAnd epsilon can be limited to be small enough, and the design of a neural network and the selection of controller parameters can ensure that sigma s plus epsilon s in the formula is more than or equal to 0, namely the requirement ofThe system is thus stable.

24) The output u of the global fast terminal sliding mode controller of the adaptive radial basis function neural network obtained in the step 22) in the formula (12)_ARBF-TSMCThe absolute value of the outer ring levitation air gap tracking controller is obtained, and the output of the outer ring levitation air gap tracking controller is made to be the reference value I of the levitation current_Aref：

25) The reference value I of the levitation current_ArefMinus the actual value I thereof_fTo obtain its error e₃(t), namely: e.g. of the type₃(t)＝I_Aref-I_f(ii) a According to the finite time control theory, the reference value of the voltage U of the levitation winding is designed as follows:

in the formula, λ₂∈(0,1)，k₃>0，k₄>0，λ₂、k₃、k₄Are all controller adjustable parameters; tanh (·) represents a hyperbolic tangent function;

equation (22) constitutes a continuous finite time controller as the inner loop levitation current tracking controller.

26) Tracking the output u of the inner loop suspension current tracking controller_AC-FTCSending the current to a PWM module to generate a driving signal of the suspension converter so as to control the suspension current I_fAnd keeping the rotating body stably suspended at the balance point.

The invention has the beneficial effects that: the invention adopts a self-adaptive Radial Basis Function (RBF) neural network finite time control strategy, on one hand, the finite time control strategy is adopted, the rapid convergence of a magnetic suspension system in finite time is ensured, the rapid tracking capability and stability of the system are ensured, the robustness of the system is improved, and no matter an outer ring global rapid terminal sliding mode controller or an inner ring continuous finite time controller, a continuous control law is output, so that the buffeting problem caused by the intrinsic non-continuous characteristic of a symbolic function item in the traditional terminal sliding mode controller and a general finite time controller is eliminated; on the other hand, as the actual working environment of the wind turbine generator is complex and changeable, and uncertain unknown time-varying interference caused by randomness of wind speed and wind direction is solved, an adaptive RBF neural network model is utilized, uncertain unknown items of a system are estimated on line in real time, the anti-interference capacity of the system is enhanced, the rotating body of the magnetic suspension vertical axis wind turbine generator is ensured to be stably suspended at a balance point in the power generation process of the wind turbine generator, and the wind turbine generator is enabled to run safely and reliably.

Drawings

FIG. 1 is a schematic structural diagram of a magnetic suspension vertical axis wind turbine according to the present invention.

Fig. 2 is a schematic diagram of a suspension system structure and a mechanical analysis of the magnetic suspension disk type motor.

Fig. 3 is a block diagram of a levitation control system based on finite time control according to the present invention.

FIG. 4 is a block diagram of a floating control system based on adaptive RBF neural network finite time control according to the present invention.

FIG. 5 is a model structure of RBF neural network of the present invention.

FIG. 6 is a block diagram of a conventional non-time-limited dual-loop PID controlled levitation control system.

Fig. 7 is a structural block diagram of a levitation control system adopting a global fast terminal sliding mode control-continuous finite time control strategy (TSMC-CFTC).

Fig. 8 is a graph of the variation of the determined disturbance imposed by the present invention.

Fig. 9 is a graph showing the variation of the uncertain disturbance applied by the present invention.

FIG. 10 is a graph of a comparison simulation of a suspended air gap under the action of an interfering force according to the present invention and a conventional non-time-limited dual-loop PID control strategy.

FIG. 11 is a graph of comparative simulation of levitation current under the action of disturbance force according to the present invention and the conventional non-time-limited dual-loop PID control strategy.

FIG. 12 is a graph of a comparison simulation of the floating air gap under the influence of interference for the finite time control strategy of the present invention without an adaptive RBF neural network estimator.

FIG. 13 is a graph of levitation current versus simulation under disturbance for the finite time control strategy of the present invention without an adaptive RBF neural network estimator.

Reference numbers in the figures: the wind power generation system comprises a 1-permanent magnet direct-drive type wind power generator, a 11-permanent magnet direct-drive type wind power generator stator, a 12-permanent magnet direct-drive type wind power generator rotor, a 2-magnetic suspension disc type motor, a 21-disc type stator, a 22-disc type rotor, a 3-wind wheel, a 4-air gap sensor, a 5-rotating shaft, a 6-supporting frame, an 18-suspension converter, a 211-disc type iron core, a 212-suspension winding, a 221-disc type rotor iron core, a 222-disc type rotor winding, a 30-suspension controller, a 31-outer ring suspension air gap tracking controller and a 32-inner ring suspension current tracking controller.

Detailed Description

The present invention will be described in further detail with reference to the accompanying drawings.

As shown in fig. 1 and 2, the magnetic suspension vertical axis wind turbine generator set of the present invention includes: the magnetic suspension vertical axis wind driven generator comprises a magnetic suspension vertical axis wind driven generator, a suspension control system, a wind wheel 3, an air gap sensor 4, a rotating shaft 5, a support frame 6 and the like. The magnetic suspension vertical shaft wind driven generator consists of two motors, namely: a permanent magnet direct drive type wind driven generator 1 and a magnetic suspension disk type motor 2.

The permanent magnet direct drive type wind driven generator 1 comprises a stator 11 and a rotor 12; the magnetic suspension disc type motor 2 is positioned below the permanent magnet direct-drive type wind driven generator 1 and comprises a disc type stator 21 and a disc type rotor 22, and the distance between the disc type stator 21 and the disc type rotor 22 is a suspension air gap delta; the disc stator 21 consists of a disc stator core 211 and a suspension winding 212, the suspension winding 212 is a direct-current excitation winding, and the air gap sensor 4 is attached to the surface of the disc stator core 211 to measure a suspension air gap; the disc rotor 22 includes a disc rotor core 221 and a disc rotor winding 222, and the disc rotor winding 222 is a three-phase winding.

As shown in fig. 1, the rotor 12 of the permanent magnet direct drive type wind turbine generator 1, the disc rotor 22 of the magnetic suspension disc motor 2, and the wind wheel 3 are all fixed to the rotating shaft 5, and these rotating parts are collectively referred to as a rotating body.

As shown in fig. 3, 4, 6 and 7, the levitation control system is composed of a levitation current transformer 18 and a levitation controller 30. The levitation current transformer 18 is a DC/DC current transformer, and is connected to the levitation winding 212 to directly control and regulate the current (i.e., levitation current) of the levitation winding 212; the suspension controller 30 is composed of an outer ring suspension air gap tracking controller 31 and an inner ring suspension current tracking controller 32, and respectively realizes the tracking control of the suspension air gap and the suspension current.

The invention relates to a suspension control method of a vertical axis wind turbine generator based on adaptive neural network finite time control, which comprises the following steps:

step 1, when the wind speed reaches the cut-in wind speed of the magnetic suspension vertical axis wind turbine, as shown in fig. 3, the reference value delta of the suspension air gap_refBy difference with its actual measured value delta (measured by the air gap sensor 4, the same applies hereinafter), its error e₁Obtaining a reference value I of the current (hereinafter referred to as "suspension current") of the suspension winding 212 through the action of an outer ring suspension air gap tracking controller 31 adopting a global fast terminal sliding mode control strategy_ref. Then, adding I_refAnd its actual value of the levitation current I_fAnd (3) performing difference, namely, sending the difference into a PWM module after the action of an inner-ring suspension current tracking controller 32 adopting a continuous finite time control strategy, generating a driving signal of the suspension converter 18, controlling the output current of the suspension converter 18, namely, controlling the suspension current, so that the rotating body stably and upwards suspends according to a preset reference track, and realizing stable and reliable operation at a suspension balance point. At the moment, no friction force exists between the rotating body of the wind turbine generator and the support frame 6, and low-wind-speed starting can be realized. The specific method comprises the following steps:

11) designing an outer ring global fast terminal sliding mode controller:

A1. designing a global fast terminal sliding mode surface as follows:

in the formula, e₁To be suspendedAir gap tracking error: e.g. of the type₁＝δ_ref-δ，δ_refIs a suspended air gap reference value, and delta is a suspended air gap measured value; alpha is alpha₀、β₀>0，p₀And q is₀Is a positive odd number, and p₀>q₀。

The derivation for equation (1) is:

in the formula (I), the compound is shown in the specification,andare respectively delta and delta_refSecond derivative with respect to time t.

As shown in fig. 2, according to the upward levitation suction force applied to the rotating body in the axial direction, the downward self-gravity of the rotating body and the external disturbance force, the kinetic equation of the rotating body in the vertical direction can be obtained as follows:

wherein m is the mass of the rotating body, and g is the gravity acceleration; f. of_d(t) is unknown time-varying interference; k is mu₀N²S/4, wherein₀The magnetic permeability is vacuum magnetic permeability, S is the effective area of the magnetic pole surface of the disc stator core 211, and N is the number of turns of the suspension winding 212; i is_fIs the current suspending winding 212, i.e., the levitation current.

From formula (3):

wherein d (t) ═ f_d(t)/m，f(x)＝-k/(mδ²)，d (t), f (x), u (t) respectively represent the system uncertain item, the system known item and the output of the terminal sliding mode controller.

When formula (4) is substituted for formula (2), it is possible to obtain:

A2. and (3) obtaining the output of the global fast terminal sliding mode controller:

the terminal sliding mode approximation law is as follows:

in the formula (I), the compound is shown in the specification,η>0, p and q are positive odd numbers, and p>q。

Comparing formula (5) with formula (6), and using u_TSMCAnd replacing u (t), obtaining the output of the global fast terminal sliding mode controller as follows:

and (3) stability analysis:

constructing a Lyapunov function:

the derivation of equation (23) is:

according to the Lyapunov stability theory, the global fast terminal sliding mode controller can enable an outer ring system to be stable.

Finite time analysis of sliding mode arrival stage and sliding stage:

setting the time from s (0) ≠ 0 to s ═ 0 as t at the sliding mode arrival stage_r1Equation (6) is modified by the equation to obtain:

the equation (24) is solved to obtain the time t of the sliding mode reaching stage_r1Comprises the following steps:

set at the sliding stage by₁(0) Not equal to 0 converges to equilibrium state e₁Time t is 0_s1When the system reaches the sliding mode surface, the following steps are provided:

the formula (25) is modified by the formula:

the equation (26) is solved to obtain the time t of the sliding stage_s1Comprises the following steps:

from this it can be concluded that: for the deviation state occurring at any time of the air gap outer ring of the magnetic suspension system, the parameters of the controller are setEta, p, q and alpha₀、β₀、p₀、q₀The global fast terminal sliding mode control strategyCan be in a limited time (t)_r1+t_s1) Internally converging to an equilibrium point.

A3. The output u of the global fast terminal sliding mode controller obtained in the formula (7) in the step A2 is processed_TSMCAfter the absolute value of the current is squared, the output of the outer ring levitation air gap tracking controller 31 is obtained and is made to be the reference value I of the levitation current_ref：

12) The inner loop levitation current tracking controller 32 is designed using a continuous finite time control strategy:

according to the principles of electromagnetism, the voltage equation for the levitation winding 212 can be derived as:

where U is the input voltage to the levitation winding 212 and R_fResistance of the levitation winding 212; d delta/dt is the first derivative of the suspended air gap delta with respect to time t; dI_fDt is the current I of the levitation winding 212_fThe first derivative over time t.

Then there are:

the reference value I of the suspension current obtained in the formula (8) in the step A3 is_refMinus the actual value I thereof_fTo obtain its error e₂(t), namely: e.g. of the type₂(t)＝I_ref-I_f(ii) a According to the finite time control theory, the reference value of the voltage U of the levitation winding is designed as:

in the formula, λ₁∈(0,1)，k₁>0，k₂>0，λ₁、k₁、k₂Are all controller adjustable parameters; tanh (·) represents a hyperbolic tangent function;

equation (9) constitutes a continuous finite time controller as the inner loop levitation current tracking controller 32.

And (3) stability analysis:

constructing a Lyapunov function:

the derivative of the formula (28) is substituted with the formula (27) or the formula (9), and u is used_C-FTCU in the alternative (27) is:

selecting appropriate parameters such that k₂Satisfy the requirement ofAnd isEquation (29) can be rewritten as:

according to Lyapunov stability theory, the continuous finite time controller can enable the inner ring system to be stable.

Finite time analysis:

from formula (30):

from the formula (28), it can be seen that V is continuously differentiable₁₂Positive definite, and there are positive real numbersAndsatisfies the following conditions:

according to the Lyapunov finite time stability theory, the corresponding convergence time T can be obtained₁Comprises the following steps:

equation (32) shows that for a deviation state e occurring at any time in the current inner loop of the magnetic levitation system₂(T) ≠ 0, and the system can always be in a limited time T₁Internally converging to the origin.

13) The output of the inner loop levitation current tracking controller 32 is sent to the PWM module to generate a driving signal for the levitation current transformer 18 to control the output current of the levitation current transformer 18, i.e. to control the current I to the levitation winding 212_fThe rotating body of the wind turbine generator is suspended upwards, and the rotating body is ensured to be stabilized at a balance point.

Step 2, after stable suspension is realized, as shown in fig. 4, the outer-loop suspended air gap tracking controller 31 uses the adaptive RBF neural network global fast terminal sliding mode control strategy instead, and the inner-loop suspended current tracking controller 32 still uses the continuous finite time control strategy to control the suspended current I_fAnd the rotating body is kept stably suspended at the suspension balance point. At the moment, the system has stronger robustness and anti-interference capability based on the estimation effect of the self-adaptive RBF neural network on the uncertain unknown items of the system, and the specific method comprises the following steps:

21) the estimation of d (t) is realized by using a system uncertain unknown item d (t) in an RBF neural network approximation formula (7), and the specific method comprises the following steps:

B1. determining the structure of the RBF neural network:

as shown in fig. 5, RBF godThe network is composed of 1 input layer, 1 hidden layer and 1 output layer, the input layer comprises 2 neurons, and the corresponding input vector is:is e₁The derivative with respect to time t; the hidden layer includes 5 neurons (n-5); the output layer includes 1 neuron.

B2. And selecting a Gaussian function as an activation function of each neuron of the hidden layer, wherein the output of the hidden layer is as follows:

in the formula, h_jOutput of the jth neuron of the hidden layer, c_j＝[c_j1,c_j2]^TIs the central vector of the j-th neuron Gaussian basis function of the hidden layer, | | E-c_jI is the Euclidean norm measuring the input vector E and the j-th neuron central vector of the hidden layer, b_jIs the width vector of the gaussian basis function of the jth neuron of the hidden layer.

B3. Determining the output of the RBF neural network:

calculating the output of the RBF neural network by using the weighted sum of the output value of the hidden layer and the weight from the hidden layer to the output layer, and enabling the output of the output layer to be the estimated value of the system uncertain unknown item d (t) in the formula (7)Then there are:

in the formula (I), the compound is shown in the specification,weight vector, w, representing the output layer_j(j ═ 1,2, …,5) are weights for the hidden layer jth neuron to the output layer neuron; h ═ h₁,h₂,h₃,h₄,h₅]^TAn output vector, h, representing the hidden layer_j(j ═ 1,2, …,5) is the output of the jth neuron in the hidden layer, and is obtained by equation (10).

22) Obtaining the output of the global fast terminal sliding mode controller of the self-adaptive RBF neural network:

using the output of the RBF neural network according to equation (7)Replacing an uncertain unknown item d (t) in a global fast terminal sliding mode controller, and using u_ARBF-TSMCSubstitution u_TSMCThe output of the global fast terminal sliding mode controller of the self-adaptive RBF neural network can be obtained as follows:

in the formula (I), the compound is shown in the specification,obtained from the formula (33).

23) Determining the self-adaptive law of RBF neural network weight updating:

C1. let the system uncertain element d (t) be represented as:

d(t)＝w^*Th+ε (13)

in the formula, w^*The ideal weight of the output layer of the RBF neural network is obtained; epsilon is an approximation error of the RBF neural network, and the error can be limited to be small enough based on the infinite precision approximation function of the RBF neural network, and epsilon is less than or equal to | epsilon |_N，ε_NIs the smallest supremum of epsilon and is a bounded, positive real number.

According to equations (11) and (13), the approximation error of the system uncertainty unknown term d (t) can be expressed as:

in the formula (I), the compound is shown in the specification,the deviation of the network weight value is obtained; under the action of the RBF neural network, the bounded real number sigma is not less than 0, so that the approximation error meets the requirement

By u_ARBF-TSMCIn place of u (t), formula (12) and formula (13) are substituted for formula (5), and the corresponding sliding mode approach law is obtained by combining formula (14):

C2. constructing a Lyapunov function:

wherein γ > 0.

Derivation of equation (16) and substitution of equation (15) gives:

C3. taking the adaptive law of updating the weight of the radial basis function neural network as follows:

by substituting formula (18) for formula (17), there are

Let η be σ/| s^q/pZeta + zeta >0, then

Since q and p are both positive odd numbers, in formula (20)If true; approximation error satisfaction of uncertain unknown items of the system due to radial basis function neural networkAnd epsilon can be limited to be small enough, and the design of a neural network and the selection of controller parameters can ensure that sigma s plus epsilon s in the formula is more than or equal to 0, namely the requirement ofThe system is thus stable.

Finite time analysis of sliding mode arrival stage and sliding stage:

setting the time from s (0) ≠ 0 to s ═ 0 as t at the sliding mode arrival stage_r2According to equation (15), then:

order toThe sliding mode approach law can be rewritten as:

the formula (34) is modified by the formula:

the time t of the sliding mode reaching stage can be obtained by solving the formula (35)_r2Comprises the following steps:

due to the fact that

Therefore, it isThe time t of the sliding mode arrival phase_r2Satisfies the following conditions:

similar to the time course of the sliding phase in step 1, the sliding phase can be obtained by e₁(0) Not equal to 0 converges to equilibrium state e₁Time t of 0_s2Comprises the following steps:

from this it can be concluded that: for the deviation state occurring at any time of the air gap outer ring of the magnetic suspension system, the parameters of the controller are setZeta, p, q and alpha₀、β₀、p₀、q₀The global fast terminal sliding mode control strategy can be always in limited time (t)_r2+t_s2) Internally converging to an equilibrium point.

24) The output u of the global fast terminal sliding mode controller of the self-adaptive RBF neural network obtained in the step 22) in the formula (12)_ARBF-TSMCThe absolute value of the current is squared to obtain the output of the outer-loop levitation air gap tracking controller 31, which is taken as the reference value I of the levitation current_Aref：

25) Reference value I of the levitation current_ArefMinus the actual value I thereof_fTo obtain its error e₃(t), namely: e.g. of the type₃(t)＝I_Aref-I_f(ii) a According to the finite time control theory, the reference value of the voltage U of the levitation winding 212 is designed as:

in the formula, λ₂∈(0,1)，k₃>0，k₄>0，λ₂、k₃、k₄Are all controller adjustable parameters; tanh (·) represents a hyperbolic tangent function;

equation (22) constitutes a continuous finite time controller as the inner loop levitation current tracking controller 32.

Stability and limited time analysis:

constructing a Lyapunov function:

the derivative of the formula (36) is substituted with the formula (27) or the formula (22), and u is used_AC-FTCU in the alternative (27) is:

selecting appropriate parameters such that k₄Satisfy the requirement ofAnd isEquation (37) can be rewritten as:

according to the Lyapunov stability theory, the design of the continuous finite time controller can stabilize the inner ring system, and the corresponding convergence time T is₂Comprises the following steps:

equation (39) shows that for a deviation state e occurring at any time in the current inner loop of the magnetic levitation system₃(T) ≠ 0, and the system can always be in a limited time T₂Internally converging to the origin.

26) Tracking the inner loop levitating current to the output u of the controller 32_AC-FTCSending the current to the PWM module to generate a driving signal of the levitation current transformer 18 so as to control the levitation winding current I_fThe rotating body is kept stably suspended at the balance point, and the rotating body can be ensured to be stabilized to the balance point within a limited time even if uncertain unknown interference is received.

The invention will be further illustrated by the following preferred embodiment.

In order to verify the effectiveness of the finite time suspension control method based on the self-adaptive RBF neural network, a double-ring PID control strategy (PID-PID for short) is respectively adopted for a suspension system of a magnetic suspension vertical axis wind turbine generator, a double-ring finite time control strategy (TSMC-CFTC for short) is adopted for an inner ring and a global fast terminal sliding mode control is adopted for an outer ring, and a comparison simulation experiment is carried out on the double-ring finite time control strategy (TSMC-ARBF-CFTC for short) of the self-adaptive RBF neural network, wherein the inner ring is controlled by continuous finite time, the outer ring is controlled by the global fast terminal sliding mode, and uncertain unknown items of the system are compensated by combining with a self-adaptive RBF neural network estimator.

As shown in fig. 6, the main ideas of the PID-PID method are: reference value delta of the levitation air gap_refThe difference between the measured value delta and the actual value delta is controlled by PID to obtain the current (i.e. the suspension current) reference value I of the suspension winding 212_ref**Then mix I_ref**And its current actual value I_fMaking a difference by PID controlThe controller is fed into the PWM module to generate a driving signal for the floating current transformer 18 to control the current I of the floating winding 212_fThe rotating body is suspended upward, and stable operation at the equilibrium point is finally achieved.

As shown in fig. 7, the main ideas of the TSMC-CFTC method are: reference value delta of the levitation air gap_refThe difference between the measured value delta and the actual value delta is obtained by the outer-loop floating air gap tracking controller 31 to obtain the current (i.e. the floating current) reference value I of the floating winding 212_refThen mix I_refAnd its current actual value I_fMaking difference, sending the difference to PWM module via continuous finite time controller to generate driving signal of the suspension converter 18 to control the suspension current I_fThe rotating body is suspended upward, and stable operation at the equilibrium point is finally achieved.

The specific model parameters in the simulation are shown in table 1.

TABLE 1 model parameters of magnetic levitation vertical axis wind turbine

The specific controller parameters in the simulation are shown in table 2:

TABLE 2 parameters of the levitation controller

In order to verify the anti-interference capability of the suspension system, determined interference and uncertain interference are simultaneously applied to the system, wherein the determined interference mainly comprises periodic interference and non-periodic interference, and the uncertain interference is random interference. As shown in FIG. 8 and FIG. 9, the superposition of periodic interference and random interference is applied to the system within 10s to 20s, if the wind driven generator is provided with five blades, the periodic interference with the maximum amplitude of 980N is to be applied to each blade, and the random interference takes a random value between-500N and 500N. When simulating the action of gust in 25-30 s, the interference is constant at 980N. After 30-35 s of wind gust simulation or under various sudden abnormal conditions, the non-periodic interference of oscillation is attenuated by the natural frequency from the initial constant interference force 980N.

The simulation results are shown in fig. 10, 11, 12 and 13, respectively.

FIGS. 10 and 11 are respectively the floating air gap and floating current variation curves of PID-PID control under the influence of disturbance and the TSMC-ARBF-CFTC control strategy of the present invention. Through comparative analysis, although the two control strategies can enable the system to reach an expected balance point, the PID-PID control strategy has an obvious overshoot phenomenon in a starting section, so that a dangerous impact problem is brought to the actual system, and hardware abrasion is caused; when the interference exists, the PID-PID control strategy can only ensure that the tracking error of the suspended air gap fluctuates within the range of 0.4mm, and the TSMC-ARBF-CFTC control strategy can ensure that the tracking error of the suspended air gap fluctuates within the range of 0.1mm, so that the tracking error is reduced by four times compared with the PID-PID control strategy. Therefore, the TSMC-ARBF-CFTC control strategy has a more stable starting process and stronger anti-interference capability near a balance point.

FIGS. 12 and 13 are floating air gap and floating current curves for the TSMC-CFTC control under disturbance and the TSMC-ARBF-CFTC control strategy of the present invention, respectively. Through comparative analysis, the two control strategies can enable the system to stably reach an expected balance point in the starting stage; but in terms of anti-interference capability, the TSMC-CFTC control strategy can ensure that the tracking error of the suspended air gap fluctuates within the range of 0.3mm, and after the TSMC-ARBF-CFTC control strategy is added into the adaptive RBF neural network interference estimator, the fluctuation range of the suspended air gap tracking error is reduced by 0.2mm compared with the TSMC-CFTC control strategy without the interference estimator. Particularly in the stage of existence of constant interference, the TSMC-ARBF-CFTC control strategy can ensure that the tracking error of the suspended air gap is converged to a balance point from 0.05mm within 0.6s, and the TSMC-CFTC control strategy without an interference estimator always has a steady-state error of about 0.1mm in the existence period of the constant interference. Therefore, the TSMC-ARBF-CFTC control strategy can improve the dynamic response speed and the anti-interference capability of the suspension system of the magnetic suspension vertical axis wind turbine generator.

In a word, the self-adaptive RBF neural network finite time control-based suspension control strategy has strong self-adaptive capacity, can realize stable starting and stable operation at a balance point of the magnetic suspension vertical axis wind turbine generator, and has higher dynamic response speed, stronger stability and anti-interference capacity.