CN113090461A - Low wind speed vertical axis wind turbine suspension control method based on sliding mode neural network model prediction - Google Patents
Low wind speed vertical axis wind turbine suspension control method based on sliding mode neural network model prediction Download PDFInfo
- Publication number
- CN113090461A CN113090461A CN202110519512.9A CN202110519512A CN113090461A CN 113090461 A CN113090461 A CN 113090461A CN 202110519512 A CN202110519512 A CN 202110519512A CN 113090461 A CN113090461 A CN 113090461A
- Authority
- CN
- China
- Prior art keywords
- suspension
- current
- neural network
- control
- network model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 239000000725 suspension Substances 0.000 title claims abstract description 200
- 238000003062 neural network model Methods 0.000 title claims abstract description 51
- 238000000034 method Methods 0.000 title claims abstract description 39
- 238000013528 artificial neural network Methods 0.000 claims abstract description 15
- 238000011217 control strategy Methods 0.000 claims abstract description 12
- 238000005457 optimization Methods 0.000 claims abstract description 9
- 238000005096 rolling process Methods 0.000 claims abstract description 9
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 6
- 238000005339 levitation Methods 0.000 claims description 56
- 238000004804 winding Methods 0.000 claims description 54
- 210000002569 neuron Anatomy 0.000 claims description 15
- 239000011159 matrix material Substances 0.000 claims description 12
- 238000013178 mathematical model Methods 0.000 claims description 11
- 238000012549 training Methods 0.000 claims description 11
- 239000013598 vector Substances 0.000 claims description 9
- 238000009795 derivation Methods 0.000 claims description 7
- XEEYBQQBJWHFJM-UHFFFAOYSA-N Iron Chemical group [Fe] XEEYBQQBJWHFJM-UHFFFAOYSA-N 0.000 claims description 4
- NAWXUBYGYWOOIX-SFHVURJKSA-N (2s)-2-[[4-[2-(2,4-diaminoquinazolin-6-yl)ethyl]benzoyl]amino]-4-methylidenepentanedioic acid Chemical compound C1=CC2=NC(N)=NC(N)=C2C=C1CCC1=CC=C(C(=O)N[C@@H](CC(=C)C(O)=O)C(O)=O)C=C1 NAWXUBYGYWOOIX-SFHVURJKSA-N 0.000 claims description 3
- 101000802640 Homo sapiens Lactosylceramide 4-alpha-galactosyltransferase Proteins 0.000 claims description 3
- 102100035838 Lactosylceramide 4-alpha-galactosyltransferase Human genes 0.000 claims description 3
- 230000005284 excitation Effects 0.000 claims description 3
- 230000004907 flux Effects 0.000 claims description 3
- 230000005484 gravity Effects 0.000 claims description 2
- 150000001875 compounds Chemical class 0.000 claims 1
- 230000008878 coupling Effects 0.000 abstract description 4
- 238000010168 coupling process Methods 0.000 abstract description 4
- 238000005859 coupling reaction Methods 0.000 abstract description 4
- 238000012545 processing Methods 0.000 abstract description 4
- 238000005265 energy consumption Methods 0.000 abstract description 3
- 230000035772 mutation Effects 0.000 abstract 1
- 238000007667 floating Methods 0.000 description 8
- 238000004088 simulation Methods 0.000 description 7
- 238000010586 diagram Methods 0.000 description 5
- 230000008569 process Effects 0.000 description 5
- 238000004458 analytical method Methods 0.000 description 4
- 230000004044 response Effects 0.000 description 4
- 238000013016 damping Methods 0.000 description 3
- 238000011161 development Methods 0.000 description 3
- 238000013459 approach Methods 0.000 description 2
- 238000009530 blood pressure measurement Methods 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 230000000052 comparative effect Effects 0.000 description 2
- 230000001276 controlling effect Effects 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 230000035699 permeability Effects 0.000 description 2
- 238000010248 power generation Methods 0.000 description 2
- 230000003044 adaptive effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000037396 body weight Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000009434 installation Methods 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 230000001105 regulatory effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
Images
Classifications
-
- F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
- F03—MACHINES OR ENGINES FOR LIQUIDS; WIND, SPRING, OR WEIGHT MOTORS; PRODUCING MECHANICAL POWER OR A REACTIVE PROPULSIVE THRUST, NOT OTHERWISE PROVIDED FOR
- F03D—WIND MOTORS
- F03D7/00—Controlling wind motors
- F03D7/06—Controlling wind motors the wind motors having rotation axis substantially perpendicular to the air flow entering the rotor
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
- F05—INDEXING SCHEMES RELATING TO ENGINES OR PUMPS IN VARIOUS SUBCLASSES OF CLASSES F01-F04
- F05B—INDEXING SCHEME RELATING TO WIND, SPRING, WEIGHT, INERTIA OR LIKE MOTORS, TO MACHINES OR ENGINES FOR LIQUIDS COVERED BY SUBCLASSES F03B, F03D AND F03G
- F05B2270/00—Control
- F05B2270/70—Type of control algorithm
- F05B2270/709—Type of control algorithm with neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
- G06F2119/02—Reliability analysis or reliability optimisation; Failure analysis, e.g. worst case scenario performance, failure mode and effects analysis [FMEA]
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
- Y02E10/00—Energy generation through renewable energy sources
- Y02E10/70—Wind energy
- Y02E10/74—Wind turbines with rotation axis perpendicular to the wind direction
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computer Hardware Design (AREA)
- Geometry (AREA)
- Medical Informatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- Software Systems (AREA)
- Sustainable Development (AREA)
- Sustainable Energy (AREA)
- Chemical & Material Sciences (AREA)
- Combustion & Propulsion (AREA)
- Mechanical Engineering (AREA)
- Control Of Eletrric Generators (AREA)
- Feedback Control In General (AREA)
Abstract
The invention relates to a suspension control method of a low-wind-speed vertical axis wind turbine based on sliding mode neural network model prediction, and belongs to the technical field of wind power. The method combines a neural network and multi-step model predictive control to control the suspension pressure: a neural network model is adopted to approximate a suspension system of a nonlinear low-wind-speed vertical axis wind turbine, a Newton-Raphson optimization algorithm is adopted to realize rolling optimization, the optimal control quantity of a future limited time domain of the system is obtained, the deviation is timely reduced, and the actual optimal control is kept; meanwhile, the suspended current is controlled by adopting a sliding mode control strategy with an exponential approximation law as a hyperbolic tangent function, so that the quick tracking capability and the stability of the system are ensured. The method does not need to carry out linearization processing on the nonlinear magnetic suspension system, can realize multi-step prediction, has stronger anti-interference capability and lowest energy consumption, and is particularly suitable for controlling the low-wind-speed wind power suspension system with nonlinearity, strong coupling, time-varying interference such as wind speed and wind direction mutation and the like.
Description
Technical Field
The invention relates to a control method, in particular to a suspension control method of a low-wind-speed vertical axis wind turbine based on sliding mode neural network model prediction, and belongs to the technical field of wind power.
Background
Wind power with low wind speed is one of the key points of future wind power development. However, the existing high-power horizontal axis wind driven generator has the inherent defects of need of yaw to wind, large starting resisting moment, difficult control, high cost and the like, influences the healthy development of the high-power horizontal axis wind driven generator, 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 no mechanical friction, greatly reduces the starting resistance moment, can further reduce the starting wind speed, has the advantages of low starting wind speed, simple and convenient 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 (the vertical axis wind driven generator does not need wind), and is the key direction of future wind power development.
However, the magnetic levitation technology has the characteristics of high nonlinearity, strong coupling and intrinsic instability, so that the stability control is very challenging, and the current research is mostly focused on the levitation control in the fields of magnetic levitation trains, magnetic levitation bearings, magnetic levitation planar motors and the like. The linear state feedback control adopts the most suspension control strategies, but adopts a Taylor linearization method to linearize a dynamic model of a system at a balance point, so that the state feedback control is completed, the anti-interference capability and the control performance of the linear state feedback control are weak, and particularly when large disturbance exists, a linear controller designed by the method may lose stability. Therefore, compared with the traditional control strategy only considering the stability of the system, some intelligent control methods, such as sliding mode control, neural network control, fuzzy logic control and the like, are paid more attention, and can improve the overall performance of the system, such as robustness, dynamic response capability and the like. There is a neural network PID controller proposed in the literature, which utilizes a BP neural network to adjust PID controller parameters on line, but when the suspension control system suffers a large external disturbance and the suspension air gap is far from the equilibrium point, this method cannot guarantee the control performance. There is also a robust controller proposed in the literature to solve the uncertainty problem of the levitation control system, but the levitation air gap is severely limited. Some adaptive neural fuzzy controllers based on the state observer have strong control performance, but the controllers are complex in design and do not discuss the convergence speed of the system.
The Model Predictive Control (MPC) is an online optimal control algorithm with simple structure and multivariable optimization capability. However, only a few studies apply MPC to the levitation control system at present, but the MPC is subjected to linearization processing on the levitation control system, the nonlinear characteristic of the levitation system is ignored, and most MPC adopts one-step predictive control, so that the control performance is weak.
Disclosure of Invention
The main purposes of the invention are as follows: aiming at the defects and the blank of the prior art, the invention provides the suspension control method of the low-wind-speed vertical axis wind turbine, which flexibly controls suspension and adjusts rotary damping by adopting a sliding mode neural network model prediction control strategy, realizes low-wind-speed starting, improves the anti-interference performance of a suspension system, and ensures the real-time optimal suspension performance of the low-wind-speed vertical axis wind turbine.
In order to achieve the above object, the low wind speed vertical axis wind turbine of the present invention includes: the vertical axis permanent magnet direct-drive wind driven generator comprises a vertical axis permanent magnet direct-drive wind driven generator, a suspension system, a base, a pressure sensor, a wind wheel and a rotating shaft; the permanent magnet direct-drive wind driven generator comprises a stator and a rotor; the suspension system comprises a magnetic suspension disk type motor and a suspension control system.
The magnetic suspension disc type motor is positioned below the vertical axis permanent magnet direct-drive 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 thereof, the suspension converter is connected with the suspension winding, and a main control chip of the suspension converter is a DSP (digital signal processor); the suspension controller comprises an outer ring suspension pressure tracking controller and an inner ring suspension tracking current controller; the outer ring suspension pressure tracking controller is a neural network model prediction controller, and the inner ring suspension current tracking controller is a sliding mode controller.
The wind wheel is fixed with the upper part of the rotating shaft; the disc rotor is fixed with the base, and the base is fixed with the rotating shaft. The rotor of the vertical shaft permanent magnet direct-drive type wind driven generator, the disc rotor of the magnetic suspension disc type motor, the wind wheel, the base and the rotating shaft are collectively referred to as a rotating body.
The pressure sensor is positioned right below the rotating shaft and the base and used for measuring the pressure of the rotating body acting on the base.
The invention relates to a low wind speed vertical axis wind turbine suspension control method based on sliding mode neural network model prediction, which comprises the following steps:
11) the suspension system generates an upward axial electromagnetic attraction force F after the suspension winding is electrified:
wherein i (t) is the current of the levitation winding, i.e., the levitation current; δ is the suspension air gap, i.e. the distance between the disc stator and the disc rotor; k is a radical of1=μ0N2S/4, wherein0And the magnetic pole surface effective area of the disc stator is S.
12) The resultant force of the rotating body in the vertical direction is as follows:
P=mg-F+fd(t) (2)
wherein P is the resultant force applied by the rotating body in the vertical direction, i.e. the pressure applied by the rotating body on the base; mg is the gravity of the rotating body; f. ofdAnd (t) is external disturbance force.
13) The voltage equation of the suspension winding is as follows:
where u (t) is the voltage of the floating winding, R is the resistance of the floating winding, ψ (t) is the air-gap flux linkage, L is the air-gap inductance of the floating winding, and L is 2k/δ.
14) In summary, a mathematical model of the levitation dynamics of the levitation system can be obtained:
21) constructing a suspension neural network model:
the suspension neural network model consists of an input layer, a hidden layer and an output layer.
The input layer contains two input vectors: current input i (k), current output P (k), let x1=i(k),x2P (k), the input of the input layer may be written as:
x=[i(k),P(k)]T (5)
wherein i (k) is the levitation current at the current time, p (k) is the resultant force applied to the rotating body in the vertical direction at the current time, and k is the current time.
The hidden layer contains 8 neurons, of which the input s to the jth neuron isjComprises the following steps:
in the formula, ωijAnd thetajRespectively, the connection weight and the bias vector of the hidden layer.
Output y of the jth neuron of the hidden layerjComprises the following steps:
the output layer contains 1 neuron with inputs s:
in the formula, ωjAnd θ is the connection weight and the offset vector of the output layer, respectively.
Let the output y of the output layer neurons be:
in the formula, PmAnd (k +1) is the output of the suspended neural network model at the moment k + 1.
22) Training the neural network, and outputting the suspension dynamic mathematical model output P of the suspension system and the suspension neural network model output PmThe prediction error e is P-PmAs a training signal for the suspended neural network.
And 3, transplanting the trained suspension neural network model into a DSP main control chip of the suspension converter, and establishing an actual suspension neural network model prediction control system based on the suspension converter DSP.
Step 4, designing the outer ring suspension pressure tracking controller by adopting a model predictive control strategy to realize suspension pressure tracking control, wherein the specific method comprises the following steps:
41) selecting a cost function J of the suspension system as follows:
in the formula, alpha and lambda are respectively a pressure weight factor and a current weight factor; n is a radical ofpTo predict the time domain step size, NuTo control the time domain step size, let Np=NuD is the predicted step number; p*Is a stand forA desired value of pressure of the rotating body in a vertical direction; and k is the current time.
42) The output value P of the suspended neural network model is measuredm(k) Desired value P of pressure of the rotating body in the vertical direction*(k) And the current i (k) of the levitation winding at the current moment is input into a levitation rolling optimizer; the suspension rolling optimizer adopts a Newton-Raphson (N-R) optimization algorithm to determine an optimal control input signal, namely the current optimal value i of the suspension windingopt:
iopt=i(k)-[H(k)]-1Γ(k) (11)
Wherein Γ (k) and H (k) are the Jacobian matrix and the Hessian matrix, respectively.
The first derivative is obtained for equation (10) to obtain a Jacobian matrix:
and (3) solving a second derivative of the formula (10) to obtain a Hessian matrix:
43) optimizing the current i of the suspension windingoptAnd a pressure measurement value P received by the rotating body in the vertical direction is used as an input of the suspension neural network model.
51) the current optimal value i of the suspension winding obtained in the step 4 isoptThe current i (t) of the suspension winding is subtracted to obtain a tracking error eiComprises the following steps:
ei=iopt(t)-i(t) (17)
the derivation of equation (17) is:
when formula (3) is substituted for formula (18), there are:
52) designing a sliding mode surface containing an integral term as follows:
in the formula, c1>0,c0>0。
By applying the derivation of equation (20) and substituting equation (19), the following can be obtained:
53) calculating the output of the sliding mode controller:
the exponential approximation law is:
in the formula, mu and eta are positive real numbers.
54) Substituting equation (22) into equation (21) to obtain the output of the inner-loop levitation current controller as:
55) and (3) sending the output of the inner ring suspension current controller to a PWM module, generating a driving signal of the suspension converter, thereby controlling the current i (t) of the suspension winding and keeping the rotating body in stable operation at a suspension balance point.
The invention has the beneficial effects that:
1) the outer ring suspension pressure tracking controller adopts a neural network-model predictive control strategy, combines a BP neural network with model predictive control, approaches a nonlinear suspension system through a neural network model, and does not need linearization treatment; by adopting N-R rolling optimization, the optimal control quantity of a future finite time domain of the system is obtained, a multi-step prediction function is realized, and the optimal control quantity of the future finite time domain of the system is obtained in real time, so that the problems of time variation, nonlinear interference, model mismatch and the like brought to the magnetic suspension system by the volatility and uncertainty of wind speed and wind direction are effectively solved.
2) The sliding mode controller is adopted in the inner ring suspension current controller, and the sliding mode index approximation law of the sliding mode controller adopts a smooth hyperbolic tangent function, so that the system can be stabilized in a short time, and the control precision is higher.
In a word, the invention can effectively solve the influence of nonlinear time-varying disturbance caused by wind direction random change and fan load, enhances the robustness and dynamic performance of the suspension system, has stronger anti-interference capability, ensures the real-time optimal performance of the low-wind-speed vertical-axis wind turbine, has low energy consumption and meets the low-wind-speed wind power requirement.
Drawings
FIG. 1 is a schematic structural diagram of a low wind speed vertical axis wind turbine according to the present invention.
Fig. 2 is a schematic diagram of the suspension system structure and the mechanical analysis of the invention.
FIG. 3 is a structural block diagram of a suspension control system based on sliding mode neural network model predictive control according to the present invention.
FIG. 4 is a model structure of a suspended neural network according to the present invention.
FIG. 5 is a diagram of a suspended neural network training process (system identification process) according to the present invention.
FIG. 6 is a structural block diagram of a suspension control system based on neural network-model predictive control without sliding mode control.
FIG. 7 is a graph of the suspension pressure comparison simulation of the present invention and the neural network-model predictive control based suspension pressure without sliding mode control.
FIG. 8 is a comparative simulation graph of the present invention and the suspension current based on neural network-model predictive control without sliding mode control.
Reference numbers in the figures: the wind power generation system comprises a 1-vertical axis permanent magnet direct-drive type wind power generator, an 11-vertical axis permanent magnet direct-drive type wind power generator stator, a 12-vertical axis permanent magnet direct-drive type wind power generator rotor, a 2-magnetic suspension disc type motor, a 21-magnetic suspension disc type stator, a 22-magnetic suspension disc type rotor, a 3-wind wheel, a 5-base, a 6-pressure sensor, a 7-lower end bearing, an 8-upper end bearing, a 9-shell, a 10-rotating shaft, an 11-suspension converter, a 211-disc type stator iron core, 212-suspension windings, 221-disc type rotor iron core and 222-disc type rotor windings.
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 low wind speed vertical axis wind turbine of the present invention includes: the vertical axis permanent magnet direct drive type wind driven generator comprises a vertical axis permanent magnet direct drive type wind driven generator 1, a suspension system, a wind wheel 3, a base 5, a pressure sensor 6, an upper end bearing 7, a lower end bearing 8, a shell 9 and a rotating shaft 10; the vertical axis permanent magnet direct drive type wind driven generator 1 comprises a stator 11 and a rotor 12; the suspension system comprises a magnetic suspension disk type motor 2 and a suspension control system.
The stator 11 of the vertical axis permanent magnet direct drive type wind driven generator 1 comprises a stator core and a stator winding, and the stator core is fixed with the inner side surface of the shell 9.
The magnetic suspension disc type motor 2 is positioned below the vertical axis permanent magnet direct-drive type wind driven generator 1 and comprises a disc stator 21 and a disc rotor 22; the disc stator 21 is composed of a disc stator core 211 and a suspension winding 212, and the suspension winding 212 is a direct-current excitation winding.
As shown in fig. 1, the wind wheel 3 is fixed with the upper part of the rotating shaft 10; the disc rotor 22 is fixed with the base 5, and the base 5 is fixed with the rotating shaft 10; the pressure sensor 6 is located directly below the spindle 10 and the base 5. The rotor 12 of the vertical axis permanent magnet direct drive type wind power generator 1, the disc rotor 22 of the magnetic suspension disc motor 2, the wind wheel 3, the base 5 and the rotating shaft 10 are collectively referred to as a rotating body. The pressure sensor 6 is used to measure the pressure of the rotating body acting on the base 5.
As shown in fig. 3, the levitation control system is composed of a levitation current transformer 11 and a levitation controller thereof, the levitation current transformer 11 is connected with a levitation winding 212, the levitation current transformer 11 is a DC/DC current transformer for levitation control and adjustment of rotation damping, and a main control chip thereof is a DSP (digital signal processor); the suspension controller comprises an outer ring suspension pressure tracking controller and an inner ring suspension current controller; the outer ring suspension pressure tracking controller is a neural network model prediction controller, and the inner ring suspension current controller is a sliding mode controller.
The low wind speed wind power means that the self loss of the low wind speed vertical axis wind turbine must be minimized, so the invention can realize stable suspension control and simultaneously reduce the loss of a suspension system as much as possible so as to realize low wind speed power generation. As can be seen from equation (1), the levitation force is constant, and the levitation air gap δ is proportional to the levitation current i (t), so that the energy loss of the levitation system can be reduced by reducing the levitation air gap. However, when the suspension air gap is small (2-3 mm), suspension control in the processes of floating, landing and the like is extremely difficult, so that the control target of the suspension control system is changed from the traditional suspension air gap to suspension pressure, and meanwhile, the rotation damping can be regulated and controlled by adjusting the suspension pressure, so that the working state of the low-wind-speed magnetic suspension vertical axis wind driven generator is optimal.
The invention relates to a low wind speed vertical axis wind turbine suspension control method based on sliding mode neural network model prediction control, which comprises the following steps:
11) as shown in fig. 2, the levitation system will generate an upward axial levitation force F upon energization of levitation winding 212, i.e.: the disc rotor 22 will be subjected to an upward axial suspension force F, i.e. the rotary body will be subjected to an upward axial suspension force F (because the disc rotor 22 is fixed with the rotary shaft 10 by the base 5):
where i (t) is the current of the levitation winding 212, i.e., the levitation current; δ is the suspension air gap, i.e. the distance between the disc stator 21 and the disc rotor 22; k is a radical of1=μ0N2S/4, wherein0For the vacuum permeability, N is the number of turns of the levitation winding 212, and S is the effective area of the magnetic pole surface of the disk stator 21.
12) The resultant force experienced by the rotating body in the vertical direction is:
P=mg-F+fd(t) (2)
wherein P is the resultant force applied by the rotating body in the vertical direction, i.e. the pressure applied by the rotating body on the base 5; mg is the weight of the rotating body; f. ofdAnd (t) is external disturbance force.
13) The voltage equation for the levitation winding 212 is:
where u (t) is the voltage of the floating winding 212, R is the resistance of the floating winding 212, ψ (t) is the floating air-gap flux linkage, L is the air-gap inductance of the floating winding 212, and L ═ 2k1/δ。
14) In summary, a mathematical model of the levitation dynamics of the levitation system can be obtained:
21) as shown in fig. 3 and 4, a suspended neural network model is first constructed, which uses current input values and future output values of the current suspended neural network model. The suspended neural network model consists of an input layer, a hidden layer and an output layer.
The input layer contains two input vectors: current input i (k), current output P (k), let x1=i(k),x2P (k), the input of the input layer may be written as:
x=[i(k),P(k)]T (5)
where i (k) is the levitation current at the present time, p (k) is the pressure of the rotator acting on the base 5 at the present time, and k is the present time.
The hidden layer contains 8 neurons, of which the input s to the jth neuron isjComprises the following steps:
in the formula, ωijAnd thetajRespectively, the connection weight and the bias vector of the hidden layer.
Output y of the jth neuron of the hidden layerjComprises the following steps:
the output layer contains 1 neuron with inputs s:
in the formula, ωjAnd θ is the connection weight and the offset vector of the output layer, respectively.
Let the output y of the output layer neurons be:
in the formula, PmAnd (k +1) is the output of the suspended neural network model at the moment k + 1.
22) Training the neural network, as shown in fig. 3 and 5, turning off the switch S1 and placing the switch S2 in the position I; outputting P of suspension dynamic mathematical model and P of suspension neural network model of suspension systemmThe prediction error e is P-PmAs a training signal for the suspended neural network.
And 3, transplanting the trained suspended neural network model into a DSP main control chip of the suspended converter 11, and establishing an actual suspended neural network model prediction control system based on the DSP.
Step 4, designing an outer ring suspension pressure tracking controller by adopting a model predictive control strategy to realize suspension pressure tracking control, wherein the specific method comprises the following steps:
41) selecting a cost function J of the suspension system as follows:
in the formula, alpha and lambda are respectively a pressure weight factor and a suspension current weight factor; n is a radical ofpTo predict the time domain step size, NuTo control the time domain step size, let Np=NuD is the predicted step number; p*The expected value of the pressure of the rotating body in the vertical direction is obtained; and k is the current time.
42) As shown in fig. 3, switch S1Closure, S2Setting the output value P of the suspended neural network model at the position IIm(k) Expected value P of pressure of rotating body in vertical direction*(k) And the measured value i (k) of the levitation current at the current moment is input to the levitation rolling optimizer. The suspension rolling optimizer adopts a Newton-Raphson (N-R) optimization algorithm to determine the optimal control input signal, namely the current optimal value i of the suspension winding 212opt. The iterative formula of the N-R optimization algorithm can be expressed as:
i(k+1)=i(k)-[H(k)]-1Γ(k)
wherein i (k +1) is ioptThen, the above formula can be rewritten as:
iopt=i(k)-[H(k)]-1Γ(k) (11)
wherein Γ (k) and H (k) are the Jacobian matrix and the Hessian matrix, respectively.
The first derivative is obtained for equation (10) to obtain a Jacobian matrix:
and (3) solving a second derivative of the formula (10) to obtain a Hessian matrix:
at present, model predictive control is mostly performed by adopting one-step predictive control, and the control performance of a suspension system is continuously improved along with the increase of the predicted step number d of the model predictive control. However, in order to increase the dynamic response speed, the calculation cost must be reduced, and for this purpose, two-step predictive control is adopted, i.e., d is 2. In this case, equations (10), (12) and (13) can be rewritten as follows:
by substituting formulae (14) to (16) for formula (11), i can be obtainedopt。
43) Optimize the current i of the levitation winding 212optAnd the pressure measurement P of the rotating body acting on the base 5 in the vertical direction (measured by the pressure sensor 6) is used as an input of the suspended neural network model.
51) as shown in fig. 3, the optimal value i) of the current of the levitation winding 212 obtained in step 42) is calculatedoptIs subtracted from the actual value i (t) to obtain the tracking error eiComprises the following steps:
ei=iopt(t)-i(t) (17)
the derivation of equation (17) is:
when formula (3) is substituted for formula (18), there are:
52) designing a sliding mode surface containing an integral term as follows:
in the formula, c1>0,c0>0。
By applying the derivation of equation (20) and substituting equation (19), the following can be obtained:
53) calculating the output of the sliding mode controller:
the exponential approximation law is:
in the formula, mu and eta are positive real numbers.
The sliding mode controller is demonstrated below to be globally asymptotically stable.
The Lyapunov function is constructed as:
the derivation of the above formula is:
according to the Lyapunov stability theory, the sliding mode controller is proved to be globally and gradually stable.
54) When equation (22) is substituted into equation (21), the output of the inner-loop levitation current controller is obtained as:
55) and the output of the inner-ring suspension current tracking controller is sent to a PWM module to generate a driving signal of the suspension converter 11, so that the current i (t) of the suspension winding 212 is controlled, and the rotating body is kept at the suspension reference point to stably operate.
The invention will be further described below with reference to a preferred embodiment.
In order to verify the effectiveness of the sliding mode neural network model prediction control method, the sliding mode neural network model prediction control strategy (hereinafter referred to as NNMPC-SMC) and the neural network model prediction strategy (hereinafter referred to as NNMPC-PID) of PID instead of sliding mode control are respectively adopted for a suspension system of a low-wind-speed vertical axis wind turbine to carry out comparative simulation analysis.
As shown in FIG. 6, the main ideas of the NNMPC-PID method are: the outer loop levitation pressure is tracked to the controller output (i.e., the current optimum i of levitation winding 212)opt) And the difference is made with the current value i (t) of the suspension winding 212, and the difference is sent to a PWM module through a PID controller to generate a driving signal of the suspension converter 11, so as to control the current i (t) of the suspension winding 212 and keep the rotating body in stable operation at the suspension reference point.
The specific simulation parameters are shown in table 1.
TABLE 1 model parameters of low wind speed and low wind speed vertical axis wind turbine
Parameter name | Numerical value |
Rotating body weight m (kg) | 25.5 |
Exciting winding turn number N of suspension electromagnet | 40 |
Magnetic pole area S (m) of suspension electromagnet2) | 1.2×10-2 |
Suspension air gap delta (mm) | 3 |
Desired value of levitation pressure P*(N) | 50 |
Exciting winding resistance R (omega) of suspension electromagnet | 0.7 |
Magnetic permeability mu in vacuum0(N/A2) | 4π×10-7 |
Parameters of the BP neural network are:
ωij=[0.90,0.88;0.56,0.60;0.48,0.06;0.59,0.32;0.23,0.79;0.52,0.71;0.65,0.05;0.11,1.1],θj=[-0.2;0.23;0.75;0.97;0.3;-0.72;-1.1;0.22],ωj=[-0.02;0.31;0.28;-0.11;0.26;-0.68;0.44;0.19],θ=-1.06。
the simulation results are shown in fig. 7 and 8, respectively.
FIG. 7 is a graph of suspension pressure versus simulation analysis. As can be seen from FIG. 7, in the starting stage, both control methods can be rapidly adjusted, and after the steady state is reached, the suspension pressure can be rapidly stabilized at the equilibrium point suspension pressure desired value of 50N due to the addition of the sliding mode controller, while the NNMPC-PID method without the addition of the sliding mode controller has larger suspension pressure fluctuation which exceeds 5N at most. Therefore, the sliding mode controller is added, and the control precision of the suspension control system is improved.
Fig. 8 is a graph of a suspension winding current versus simulation analysis. As can be seen from FIG. 8, the response speed of the suspension current is faster, and the reaction can be quickly carried out within 0.1 s; the suspension current of the NNMPC-PID method without adding the sliding mode controller needs more than 2s of regulation time, and the dynamic response speed is slow.
As can be seen from fig. 7 and 8, the suspension control method based on the sliding mode neural network model predictive control can improve the rapidity, the accuracy and the stability of the whole system.
In a word, the BP neural network and the multi-step Model Predictive Control (MPC) are combined, the neural network model is adopted to approach the suspension system of the low-wind-speed vertical-axis wind turbine with strong coupling and nonlinearity, linearization processing is not needed, and the problems of time variation, nonlinear interference, model mismatch and the like brought to the suspension system by the volatility and uncertainty of wind speed and wind direction are effectively solved; meanwhile, considering that the N-R method has second-order convergence and high convergence speed, the N-R rolling optimization is adopted to obtain the optimal control quantity of the future limited time domain of the system, reduce the deviation in time, maintain the actual optimal control and ensure the quick tracking capability and stability of the system. Meanwhile, the suspension current controller adopts a sliding mode controller, and the sliding mode index approximation law of the suspension current controller adopts a smooth hyperbolic tangent function, so that the system can be stable in a short time, and the control precision is higher. Compared with the conventional control, the method provided by the invention does not need to carry out linearization processing on the nonlinear magnetic suspension system, can realize multi-step prediction, and can obtain the optimal control quantity of the system in a future finite time domain in real time, so that the method has stronger anti-interference capability and lowest energy consumption, and is particularly suitable for controlling the low-wind-speed wind power suspension system with nonlinearity, strong coupling and time-varying nonlinear interference.
Claims (2)
1. The suspension control method of the low-wind-speed vertical axis wind turbine based on the sliding mode neural network model prediction comprises the following steps: the vertical axis permanent magnet direct-drive wind driven generator comprises a vertical axis permanent magnet direct-drive wind driven generator, a suspension system, a base, a pressure sensor, a wind wheel and a rotating shaft; the permanent magnet direct-drive wind driven generator comprises a stator and a rotor; the suspension system comprises a magnetic suspension disk type motor and a suspension control system; the magnetic suspension disc type motor is positioned below the vertical axis permanent magnet direct-drive wind driven generator and 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 controller comprises an outer ring suspension pressure tracking controller and an inner ring suspension current tracking controller; the rotor of the vertical shaft permanent magnet direct-drive wind driven generator, the disc rotor of the magnetic suspension disc type motor, the wind wheel, the base and the rotating shaft are collectively called as a rotating body; the method is characterized by comprising the following steps:
step 1, establishing a suspension dynamic mathematical model of the suspension system:
wherein P is the resultant force applied by the rotating body in the vertical direction, i.e. the pressure applied by the rotating body on the base; mg is the gravity of the rotating body; f. ofd(t) is external disturbance force; i (t) is the current of the levitation winding, i.e. the levitation current; δ is the suspension air gap, i.e. the distance between the disc stator and the disc rotor; k is a radical of1=μ0N2S/4, wherein0The magnetic pole surface is a vacuum magnetic conductivity, N is the number of turns of the suspension winding, and S is the effective area of the magnetic pole surface of the disc stator; u (t) is the voltage of the levitation winding, R isThe resistance psi (t) of the suspension winding is a suspension air gap flux linkage, L is the air gap inductance of the suspension winding, and L is 2 k/delta;
step 2, training a suspension neural network model according to the suspension dynamic mathematical model of the suspension system, wherein the specific method comprises the following steps:
21) constructing a suspension neural network model:
the suspension neural network model consists of an input layer, a hidden layer and an output layer;
the input layer contains two input vectors: current input i (k), current output P (k), let x1=i(k),x2P (k), the input of the input layer may be written as:
x=[i(k),P(k)]T (5)
wherein i (k) is the levitation current at the current moment, p (k) is the resultant force applied to the rotating body in the vertical direction at the current moment, and k is the current moment;
the hidden layer contains 8 neurons, of which the input s to the jth neuron isjComprises the following steps:
in the formula, ωijAnd thetajRespectively are the connection weight and the offset vector of the hidden layer;
output y of the jth neuron of the hidden layerjComprises the following steps:
in the formula (f)1(. cndot.) is a hyperbolic tangent function;
the output layer contains 1 neuron with inputs s:
in the formula (I), the compound is shown in the specification,ωjand theta is the connection weight and the offset vector of the output layer respectively;
let the output y of the output layer neurons be:
in the formula, Pm(k +1) is the output of the suspended neural network model at the time of k + 1;
22) training the neural network, and outputting the suspension dynamic mathematical model output P of the suspension system and the suspension neural network model output PmThe prediction error e is P-PmAs a training signal for the suspended neural network;
step 3, transplanting the trained suspension neural network model into a main control chip of the suspension converter, and establishing an actual suspension neural network model prediction control system based on the main control chip of the suspension converter;
step 4, designing the outer ring suspension pressure tracking controller by adopting a model predictive control strategy to realize suspension pressure tracking control; the specific method comprises the following steps:
41) selecting a cost function J of the suspension system as follows:
in the formula, alpha and lambda are respectively a pressure weight factor and a current weight factor; n is a radical ofpTo predict the time domain step size, NuTo control the time domain step size, let Np=NuD is the predicted step number; p*The expected value of the pressure of the rotating body in the vertical direction is obtained; k is the current time;
42) the output value P of the suspended neural network model is measuredm(k) Desired value P of pressure of the rotating body in the vertical direction*(k) And the current i (k) of the levitation winding at the current moment is input into a levitation rolling optimizer; the suspension rolling optimizer adopts Newton-RapThe hson optimization algorithm determines the optimal control input signal, i.e. the optimal value i of the current of the levitation windingopt:
iopt=i(k)-[H(k)]-1Γ(k) (11)
Wherein Γ (k) and H (k) are Jacobian matrix and Hessian matrix, respectively;
the first derivative is obtained for equation (10) to obtain a Jacobian matrix:
and (3) solving a second derivative of the formula (10) to obtain a Hessian matrix:
43) optimizing the current i of the suspension windingoptAnd the pressure P of the rotator acting on the base is used as the input of the suspension neural network model;
and 5, designing the inner-ring suspended current tracking controller by adopting a sliding mode control strategy to realize suspended current tracking control.
2. The suspension control method of the low-wind-speed vertical-axis wind turbine predicted by the sliding-mode neural network model according to claim 1, characterized in that the concrete method in the step 5 is as follows:
51) the current optimal value i of the suspension winding obtained in the step 4 isoptThe current i (t) of the suspension winding is subtracted to obtain a tracking error eiComprises the following steps:
ei=iopt(t)-i(t) (17)
the derivation of equation (17) is:
substituting the second equation in equation (4), namely the voltage equation, into equation (18), there are:
52) designing a sliding mode surface containing an integral term as follows:
in the formula, c1>0,c0>0;
By applying the derivation of equation (20) and substituting equation (19), the following can be obtained:
53) calculating the output of the sliding mode controller:
the exponential approximation law is:
in the formula, mu and eta are positive real numbers;
54) substituting equation (22) into equation (21) to obtain the output of the inner-loop levitation current controller as:
55) and (3) sending the output of the inner ring suspension current tracking controller to a PWM module, and generating a driving signal of the suspension converter so as to control the current i (t) of the suspension winding and keep the rotating body in stable operation at a suspension balance point.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110519512.9A CN113090461B (en) | 2021-05-13 | 2021-05-13 | Low wind speed vertical axis wind turbine suspension control method based on sliding mode neural network model prediction |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110519512.9A CN113090461B (en) | 2021-05-13 | 2021-05-13 | Low wind speed vertical axis wind turbine suspension control method based on sliding mode neural network model prediction |
Publications (2)
Publication Number | Publication Date |
---|---|
CN113090461A true CN113090461A (en) | 2021-07-09 |
CN113090461B CN113090461B (en) | 2022-06-28 |
Family
ID=76665442
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110519512.9A Active CN113090461B (en) | 2021-05-13 | 2021-05-13 | Low wind speed vertical axis wind turbine suspension control method based on sliding mode neural network model prediction |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113090461B (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113775474A (en) * | 2021-08-21 | 2021-12-10 | 曲阜师范大学 | Suspension control method of vertical axis wind turbine generator based on self-adaptive neural network finite time control |
CN114448311A (en) * | 2022-01-24 | 2022-05-06 | 江苏大学 | Fuzzy neural network prediction decoupling control system for bearingless permanent magnet synchronous generator |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110345013A (en) * | 2019-07-24 | 2019-10-18 | 曲阜师范大学 | Magnetic suspension vertical shaft Wind turbines control method based on Neural Network model predictive control |
CN110805523A (en) * | 2019-12-13 | 2020-02-18 | 曲阜师范大学 | Suspension control method of wind power magnetic suspension yaw system based on neural network |
CN111173681A (en) * | 2020-01-09 | 2020-05-19 | 曲阜师范大学 | Magnetic suspension vertical axis wind turbine generator suspension control method based on sliding mode neural network |
CN111835261A (en) * | 2020-07-22 | 2020-10-27 | 曲阜师范大学 | Magnetic suspension vertical axis wind turbine generator suspension control method based on adaptive neural network |
-
2021
- 2021-05-13 CN CN202110519512.9A patent/CN113090461B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110345013A (en) * | 2019-07-24 | 2019-10-18 | 曲阜师范大学 | Magnetic suspension vertical shaft Wind turbines control method based on Neural Network model predictive control |
CN110805523A (en) * | 2019-12-13 | 2020-02-18 | 曲阜师范大学 | Suspension control method of wind power magnetic suspension yaw system based on neural network |
CN111173681A (en) * | 2020-01-09 | 2020-05-19 | 曲阜师范大学 | Magnetic suspension vertical axis wind turbine generator suspension control method based on sliding mode neural network |
CN111835261A (en) * | 2020-07-22 | 2020-10-27 | 曲阜师范大学 | Magnetic suspension vertical axis wind turbine generator suspension control method based on adaptive neural network |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113775474A (en) * | 2021-08-21 | 2021-12-10 | 曲阜师范大学 | Suspension control method of vertical axis wind turbine generator based on self-adaptive neural network finite time control |
CN113775474B (en) * | 2021-08-21 | 2023-09-29 | 曲阜师范大学 | Vertical axis wind turbine generator suspension control method based on adaptive neural network finite time control |
CN114448311A (en) * | 2022-01-24 | 2022-05-06 | 江苏大学 | Fuzzy neural network prediction decoupling control system for bearingless permanent magnet synchronous generator |
Also Published As
Publication number | Publication date |
---|---|
CN113090461B (en) | 2022-06-28 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Sun et al. | Reinforcement learning-based optimal tracking control for levitation system of maglev vehicle with input time delay | |
CN113090461B (en) | Low wind speed vertical axis wind turbine suspension control method based on sliding mode neural network model prediction | |
Wai et al. | Real-time PID control strategy for maglev transportation system via particle swarm optimization | |
CN107070342A (en) | A kind of control system for permanent-magnet synchronous motor of bringing onto load state observer | |
Wai et al. | On-line supervisory control design for maglev transportation system via total sliding-mode approach and particle swarm optimization | |
CN110401378B (en) | Magnetic suspension yaw motor control method based on neural network-model predictive control | |
Wai et al. | Observer-based adaptive fuzzy-neural-network control for hybrid maglev transportation system | |
CN110345013B (en) | Magnetic suspension vertical axis wind turbine generator control method based on neural network model predictive control | |
CN108183650B (en) | Wind power magnetic suspension yaw motor control method based on model prediction control | |
CN111835261B (en) | Magnetic suspension vertical axis wind turbine generator suspension control method based on adaptive neural network | |
CN111173678A (en) | Wind power cabin suspension control method for RBF neural network pitch interference compensation | |
CN110805523B (en) | Suspension control method of wind power magnetic suspension yaw system based on neural network | |
CN108488036A (en) | Wind-powered electricity generation magnetic suspension yaw system suspension control method based on model mismatch compensator | |
CN109507869A (en) | A kind of optimization method of the motor control PI parameter suitable for permanent magnet synchronous motor | |
Li et al. | Modeling of maglev yaw system of wind turbines and its robust trajectory tracking control in the levitating and landing process based on NDOB | |
Sun et al. | Neuron PID control for a BPMSM based on RBF neural network on‐line identification | |
CN111173681B (en) | Magnetic suspension vertical axis wind turbine generator suspension control method based on sliding mode neural network | |
Lin et al. | Fuzzy PID control algorithm based on PSO and application in BLDC motor | |
Ouyang et al. | An LMI-based simple robust control for load sway rejection of rotary cranes with double-pendulum effect | |
CN115473464A (en) | Magnetic suspension yaw motor control method based on neural network model predictive control | |
CN113775474B (en) | Vertical axis wind turbine generator suspension control method based on adaptive neural network finite time control | |
Chen et al. | The design of adaptive sliding mode controller based on RBFNN approximation for suspension control of MVAWT | |
CN114268263A (en) | Double-closed-loop fuzzy control method for motor driver | |
Ma et al. | Parameters self-adjusting fuzzy PID control in magnetic levitation system | |
Ding | Comparative study on control effect of permanent magnet synchronous motor based on Fuzzy PID control and BP neural network PID control |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |