CN110535396A - Surface permanent magnetic Synchronous Machine Models forecast Control Algorithm based on BP neural network - Google Patents
Surface permanent magnetic Synchronous Machine Models forecast Control Algorithm based on BP neural network Download PDFInfo
- Publication number
- CN110535396A CN110535396A CN201910817795.8A CN201910817795A CN110535396A CN 110535396 A CN110535396 A CN 110535396A CN 201910817795 A CN201910817795 A CN 201910817795A CN 110535396 A CN110535396 A CN 110535396A
- Authority
- CN
- China
- Prior art keywords
- neural network
- voltage vector
- output
- torque
- layer
- 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
Classifications
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02P—CONTROL OR REGULATION OF ELECTRIC MOTORS, ELECTRIC GENERATORS OR DYNAMO-ELECTRIC CONVERTERS; CONTROLLING TRANSFORMERS, REACTORS OR CHOKE COILS
- H02P23/00—Arrangements or methods for the control of AC motors characterised by a control method other than vector control
- H02P23/0004—Control strategies in general, e.g. linear type, e.g. P, PI, PID, using robust control
- H02P23/0018—Control strategies in general, e.g. linear type, e.g. P, PI, PID, using robust control using neural networks
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02P—CONTROL OR REGULATION OF ELECTRIC MOTORS, ELECTRIC GENERATORS OR DYNAMO-ELECTRIC CONVERTERS; CONTROLLING TRANSFORMERS, REACTORS OR CHOKE COILS
- H02P23/00—Arrangements or methods for the control of AC motors characterised by a control method other than vector control
- H02P23/04—Arrangements or methods for the control of AC motors characterised by a control method other than vector control specially adapted for damping motor oscillations, e.g. for reducing hunting
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02P—CONTROL OR REGULATION OF ELECTRIC MOTORS, ELECTRIC GENERATORS OR DYNAMO-ELECTRIC CONVERTERS; CONTROLLING TRANSFORMERS, REACTORS OR CHOKE COILS
- H02P23/00—Arrangements or methods for the control of AC motors characterised by a control method other than vector control
- H02P23/30—Direct torque control [DTC] or field acceleration method [FAM]
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02P—CONTROL OR REGULATION OF ELECTRIC MOTORS, ELECTRIC GENERATORS OR DYNAMO-ELECTRIC CONVERTERS; CONTROLLING TRANSFORMERS, REACTORS OR CHOKE COILS
- H02P27/00—Arrangements or methods for the control of AC motors characterised by the kind of supply voltage
- H02P27/04—Arrangements or methods for the control of AC motors characterised by the kind of supply voltage using variable-frequency supply voltage, e.g. inverter or converter supply voltage
- H02P27/06—Arrangements or methods for the control of AC motors characterised by the kind of supply voltage using variable-frequency supply voltage, e.g. inverter or converter supply voltage using dc to ac converters or inverters
- H02P27/08—Arrangements or methods for the control of AC motors characterised by the kind of supply voltage using variable-frequency supply voltage, e.g. inverter or converter supply voltage using dc to ac converters or inverters with pulse width modulation
- H02P27/12—Arrangements or methods for the control of AC motors characterised by the kind of supply voltage using variable-frequency supply voltage, e.g. inverter or converter supply voltage using dc to ac converters or inverters with pulse width modulation pulsing by guiding the flux vector, current vector or voltage vector on a circle or a closed curve, e.g. for direct torque control
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02P—CONTROL OR REGULATION OF ELECTRIC MOTORS, ELECTRIC GENERATORS OR DYNAMO-ELECTRIC CONVERTERS; CONTROLLING TRANSFORMERS, REACTORS OR CHOKE COILS
- H02P2207/00—Indexing scheme relating to controlling arrangements characterised by the type of motor
- H02P2207/05—Synchronous machines, e.g. with permanent magnets or DC excitation
Abstract
The invention discloses the surface permanent magnetic Synchronous Machine Models forecast Control Algorithms based on BP neural network, the present invention substitutes surface-type permanent magnet synchronous motor model prediction algorithm using BP neural network, BP neural network has the ability of powerful nonlinear fitting and pattern recognition classifier, the characteristics of its distributed parallel operation, can greatly reduce operation time and the computational burden of algorithm, the timeliness of raising system, and the accuracy rate and effect of BP neural network alternative model prediction algorithm are satisfactory, it selects the accuracy rate of optimal voltage vector to can achieve 88.34%, since neural network is not fed back and correction capability, it can effectively be solved the problems, such as by one switching value of setting and threshold value switching neural network operating mode and MPC operating mode out of control under BP neural network dynamic step.
Description
Technical field
The invention belongs to permanent magnet synchronous motor Direct Torque Control fields, and in particular to the surface-type based on BP neural network
Permanent magnet synchronous motor model predictive control method.
Background technique
Study on direct torque control technology is based on stator magnetic linkage coordinate system and directly using torque as control object, avoids rotation
A large amount of calculating when coordinate transform and the dependence to the parameter of electric machine, dynamic property is good, and the torque response time is short.But
Traditional DTC is a kind of off-line type control method, and control algolithm and good voltage vector LUT prepared in advance are implanted into micro process
In device, executed in each control loop cycle.DTC is according to the current torque error of electric machine control system and stator magnetic linkage error
Optimal voltage vector is chosen from voltage vector LUT to eliminate torque error and stator magnetic linkage error.But tradition DTC can only
Optimal voltage vector is chosen according to the motor variable error of each control period initial time, can not predict and be sweared in this voltage
Amount acts on the variation tendency of lower motor variable, not can guarantee entire control period internal torque and stator magnetic linkage remains at one
In fixed range, biggish torque pulsation and stator magnetic linkage is caused to be pulsed.Meanwhile conventional voltage vector LUT design is simple, control
Precision processed is lower, and motor can have the selective output of multiple and different voltage vectors under specific run state, is obtained according to look-up table
Voltage vector may not be optimal voltage vector.Therefore, the online control method of motor can be studied, it is pre- in real time
The motor torque error under the effect of a certain voltage vector is predicted in the variation for surveying motor variable when applying different voltages vector in advance
With variation of the stator magnetic linkage error within the entirely control period, it is ensured that the validity and accuracy of applied voltage vector.
Model Predictive Control (Model predictive control, MPC) is as a kind of effective online controlling party
Method is widely used in various Industry Control occasions.There is pertinent literature proposition to combine MPC with DTC, proposes a kind of model
Predict that Direct Torque Control, this method carry out discretization to continuous non-linear DC-motor equation and obtain motor prediction model, predict
The variation of motor torque and stator magnetic linkage in different voltages vector effect lower following multiple control periods, and according to objective function
Comprehensive assessment is carried out to each step prediction result of each voltage vector, a final determining optimal voltage vector sequence is simultaneously only defeated
The first term of sequence out.The primary goal of this method be reduce inverter switching frequency, while by motor torque, stator magnetic linkage and
Inverter neutral point voltage limits in a certain range, to guarantee that motor stabilizing is run.But the prediction algorithm proposed in document
Operand is larger, it is difficult to implement in practical applications.
Summary of the invention
The purpose of the present invention is to overcome the above shortcomings and to provide a kind of surface permanent magnetics based on BP neural network to synchronize electricity
Machine model predictive control method, it is direct by the way that the good BP neural network of off-line training is embedded into surface permanent magnetic synchronous motor
With model prediction algorithm serial design in moment controlling system, alternative model prediction algorithm is completed optimal in the following control period
The selection work of voltage vector.
In order to achieve the above object, the present invention the following steps are included:
Step 1 determines input quantity in surface permanent magnetic Synchronous Machine Models predictive control algorithm and output quantity and mind
Input quantity and output quantity through network, and determine the variation range of motor reference rotation velocity and load torque；
Step 2, according to two kinds of reference rotation velocity step under load torque step under constant reference revolving speed and constant load torque
Mode by under different reference rotation velocities load torque change the case where and different loads torque under reference rotation velocity change the case where according to
Certain step-length and interval traversal is got, and the data of the above-mentioned input quantity accordingly generated feeding surface permanent magnetic is synchronized electricity
In machine Model Predictive Control Algorithm, the optimal voltage vector sequence of Model Predictive Control Algorithm selection in the following control period is generated
Column, and by the corresponding BP neural network that is combined into of the optimal voltage vector of the value of each input quantity and corresponding selection out
Training sample；
Step 3 constructs BP neural network topological model；
Training sample data in step 2 are normalized step 4, by the training sample after normalized
This input has constructed progress off-line training study in BP neural network topological model, obtains satisfactory weight threshold；
The good BP neural network of off-line training is embedded into the prediction of surface permanent magnetic Synchronous Machine Models and directly turned by step 5
With model prediction algorithm serial design in square control system, the following control period is carried out for alternative model predictive control algorithm
The work and function of interior motor optimal voltage vector selection；
A switching value is arranged in step 6 between model prediction algoritic module and the serial design of BP neural network, when
When torque error is greater than certain threshold value, operating mode is switched to model prediction algorithm and carries out optimal voltage vector by switching value
Work is selected, when torque error value is less than the threshold value, the selection work of voltage vector is carried out using neural network.
Step 1 determines input quantity in surface permanent magnetic Synchronous Machine Models predictive control algorithm and output quantity and mind
Through network inputs amount and output quantity, the specific method is as follows:
Ignore rotor rotational movement and Stator resistance voltage dropping, after applying nonzero voltage space vector, subsequent time stator magnetic linkage width
Value and torque are as shown in formula 1.1:
It is the amplitude of current K moment stator magnetic linkage,It is the stator magnetic linkage amplitude at k+1 moment,It is current
The voltage vector magnitude to be applied, fixed amplitude set tentering value for voltage vector, amplitude is constantly equal to
Wherein UdcIt is DC bus-bar voltage；Δ t is the action time that the voltage vector applies, and α is voltage vector and stator magnetic linkage vector
Between angle；
The torque of subsequent time stator magnetic linkage is as shown in formula 1.2:
Te(k+1) be the k+1 moment motor torque, p is the number of pole-pairs of motor, ψfIt is permanent magnet flux linkage,It is current K
The amplitude of moment stator magnetic linkage, LdFor d axis stator inductance, α is the angle between voltage vector and stator magnetic linkage vector, and δ (k) is
The angle of torsion at current K moment；
Cost function used is as shown in formula 1.3:
The result that formula 1.1 and formula 1.2 are calculated is sent into cost function, in entire algorithmic procedure, if there is n it is alternative
Voltage vector can then obtain n cost function numerical value, select voltage vector corresponding to the smallest numerical value of cost function as
The calculating will be applied to the optimal voltage vector of motor in the control period；
The input quantity of Model Predictive Control Algorithm has angle of torsion, current stator magnetic linkage amplitude, current stator magnet chain angle respectively
Position, Reference Stator Flux Linkage, torque reference, alternative voltage vector angle, output quantity are selected voltage vectors.In 6 input quantities
Fall constant therein, including Reference Stator Flux Linkage and alternative voltage vector angle, characteristic dimension is simplified to 4, it is possible to reduce operation
Amount, as the input quantity of neural network, output quantity of the selected voltage vector as neural network.
It, will be upper in corresponding value range according to certain step-length and interval by programming or the Realization of Simulation in step 2
Two kinds of situation of change traversals are stated to get.
In step 3, the feedforward neural network that BP neural network topological model is three layers is public according to hidden layer node experience
FormulaDetermine node in hidden layer, wherein a is the constant between 0~10, on this basis further basis
Commissioning experience determines hidden layer node number.Input layer corresponds to 4 input quantities of above-mentioned determination, and output node layer is equal to
Number i.e. 7 basic voltage vectors of alternative voltage vector.Neuron in the hidden layer and output layer of BP neural network swashs
Function living is all made of the tanh transmission function of tansig type, the expression formula of tansig function be f (x)=2/ (1+exp (-
2*x))-1。
In step 4, normalized uses default behavior, and processing mode is as follows:
In formula, xmin is the minimum number in data series, and xmax is the maximum number in sequence.Ymin and ymax is specified
Range is normalized, is defaulted as -1 and+1.
In step 4, BP neural network topological model is trained study, and the specific method is as follows:
The first step, the training sample after normalized respectively correspond the prediction of surface permanent magnetic Synchronous Machine Models per one-dimensional
Input is combined into training sample sequence (X, Y) by dependent variable, that is, input quantity in algorithm with output, and X is input data, and Y is pair
The output data answered；
Second step, the training sample sequence (X, Y) being combined into according to system input and output determine network input layer number of nodes
N, node in hidden layer l, output layer number of nodes m, the connection weight between initialization input layer, hidden layer and output layer neuron
Value ωij, ωjk, hidden layer threshold value a, output layer threshold value b are initialized, learning rate and neuron excitation function are given；
Third step, according to input variable X, input layer and implicit interlayer connection weight ωijAnd hidden layer threshold value a, it calculates
Hidden layer exports H；
In formula, l is node in hidden layer, and f is general hidden layer excitation function, selected function are as follows:
F (x)=2/ (1+exp (- 2*x)) -1 1.6
4th step exports H, connection weight ω according to hidden layerjkWith threshold value b, BP neural network prediction output O is calculated；
5th step exports O and desired output Y according to neural network forecast, calculates neural network forecast error J；
ek=Yk-Ok 1.8
Wherein, k=1,2 ..., m；
6th step, according to J, according to gradient descent method direction calculating, and successively adjusts weight threshold by output layer.
Taking step-length is η, can obtain t+1 adjustment formula of the connection weight and threshold value between neuron u to neuron v:
7th step, judges whether algorithm iteration terminates, if iteration terminates one of the following conditions to be met, iteration terminates:
Condition 1 reaches the number of iterations 2000；
Condition 2, the error of prediction drop within target error value 0.05；
Condition 3, continuous 6 verifyings collection error do not reduce or increase instead；
If iteration is not over, third step is returned.
Compared with prior art, the present invention is calculated using BP neural network substitution surface-type permanent magnet synchronous motor model prediction
Method, BP neural network have the characteristics of ability of powerful nonlinear fitting and pattern recognition classifier, distributed parallel operation
Operation time and the computational burden that algorithm can be greatly reduced improve the timeliness of system, and the standard of BP neural network substitution
True rate and effect are satisfactory, select the accuracy rate of optimal voltage vector to can achieve 88.34%, due to neural network
It does not feed back and correction capability, passes through one switching value of setting and threshold value switches neural network operating mode and MPC operating mode
It can effectively solve the problems, such as out of control under BP neural network dynamic step, MPC can be retained under the double mode that BP is combined with MPC
Excellent performance, torque pulsation can averagely reduce 4.34%, and maximum can reduce 7.29%, and on-off times can be reduced
27.55%.The present invention has certain innovation advantage relative to traditional model prediction algorithm, demonstrates intelligent algorithm in electricity
Application prospect in machine control.
Detailed description of the invention
Fig. 1 is model flow figure of the invention；
Fig. 2 is the schematic illustration of BP neural network of the invention；
Fig. 3 is stator magnetic linkage motion change figure.
Fig. 4 is the actual motor torque figure switched under table schema；
Fig. 5 switchs the motor actual speed figure under table schema；
Fig. 6 is the stator magnetic linkage amplitude figure switched under table schema；
Fig. 7 is the stator magnetic linkage trajectory diagram switched under table schema；
Fig. 8 is a phase current waveform figure switched under table schema；
Fig. 9 is the actual motor torque figure under BP neural network mode；
Figure 10 is the motor actual speed figure under BP neural network mode；
Figure 11 is the stator magnetic linkage amplitude figure under BP neural network mode；
Figure 12 is the stator magnetic linkage trajectory diagram under BP neural network mode；
Figure 13 is a phase current waveform figure under BP neural network mode；
Figure 14 is the actual motor torque figure under MPC operating mode；
Figure 15 is the motor actual speed figure under MPC operating mode；
Figure 16 is the stator magnetic linkage amplitude figure under MPC operating mode；
Figure 17 is the stator magnetic linkage trajectory diagram under MPC operating mode；
Figure 18 is a phase current waveform figure under MPC operating mode.
Specific embodiment
The present invention will be further described with reference to the accompanying drawing.
Referring to Fig. 1, the present invention the following steps are included:
Step 1 determines input quantity in surface permanent magnetic Synchronous Machine Models predictive control algorithm and output quantity and mind
Input quantity and output quantity through network, and determine the variation range of motor reference rotation velocity and load torque；
Step 2, according to two kinds of reference rotation velocity step under load torque step under constant reference revolving speed and constant load torque
Mode by under different reference rotation velocities load torque change the case where and different loads torque under reference rotation velocity change the case where according to
Certain step-length and interval traversal is got, and the data of the above-mentioned input quantity accordingly generated feeding surface permanent magnetic is synchronized electricity
In machine Model Predictive Control Algorithm, the optimal voltage vector sequence of Model Predictive Control Algorithm selection in the following control period is generated
Column, and by the corresponding BP neural network that is combined into of the optimal voltage vector of the value of each input quantity and corresponding selection out
Training sample；
Step 3 constructs BP neural network topological model；
The training sample data being collected into step 2 are carried out data normalization processing, by normalized by step 4
Training sample data afterwards, which are input to, has constructed progress off-line training study in BP neural network topological model, is conformed to
The weight threshold asked；
The good BP neural network of off-line training is embedded into the prediction of surface permanent magnetic Synchronous Machine Models and directly turned by step 5
With model prediction algorithm serial design in square control system, alternative model predictive control algorithm carries out electric in the following control period
The work and function of machine optimal voltage vector selection；
Since neural network is not fed back and correction capability, passes through one switching value of setting and threshold value switches neural network work
Operation mode and MPC operating mode can effectively solve the problems, such as out of control under BP neural network dynamic step, and BP and MPC group
The excellent performance of MPC can be retained under the double mode of conjunction, therefore Model Predictive Control can be further substituted with.
In step 1, ignore rotor rotational movement and Stator resistance voltage dropping, after applying nonzero voltage space vector, stator magnetic linkage fortune
Dynamic variation is as shown in Figure 3.
After applying voltage vector, subsequent time stator magnetic linkage amplitude and torque are as shown in formula 1.13 and formula 1.14.
Wherein, It is the amplitude of current K moment stator magnetic linkage,It is determining for k+1 moment
Sub- magnetic linkage amplitude,It is the voltage vector magnitude currently to be applied, fixed amplitude sets tentering value for voltage vector,
Its amplitude is constantly equal toWherein UdcIt is DC bus-bar voltage；Δ t is the action time that the voltage vector applies, and α is
Angle between voltage vector and stator magnetic linkage vector.
Te(k+1) be the k+1 moment motor torque, p is the number of pole-pairs of motor, ψfIt is permanent magnet flux linkage,It is current K
The amplitude of moment stator magnetic linkage, LdFor d axis stator inductance, α is the angle between voltage vector and stator magnetic linkage vector, and δ (k) is
The angle of torsion at current K moment.
Cost function used is as shown in formula 1.15:
The result that formula 1.13 and formula 1.14 are calculated is sent into cost function, in entire algorithmic procedure, if there is n it is standby
Voltage vector is selected, then can obtain n cost function numerical value, voltage vector corresponding to the smallest numerical value of cost function is selected to make
The optimal voltage vector that motor is applied in the period is controlled for the calculating.
Model Predictive Control Algorithm can be learnt in formula by above-mentioned surface permanent magnetic Synchronous Machine Models PREDICTIVE CONTROL
Input quantity have respectively angle of torsion, current stator magnetic linkage amplitude, current stator magnetic linkage Angle Position, Reference Stator Flux Linkage, torque reference,
Alternative voltage vector removes constant therein, because constant has little significance in the operation of neural network, including Reference Stator Flux Linkage
With alternative voltage vector angle, it is believed that the selection of current voltage vector is mainly related with current stator magnetic linkage Angle Position.Therefore,
Characteristic dimension is simplified to 4, it is possible to reduce operand, using 4 variables as the input of neural network, selected voltage arrow
Measure the output as neural network.
In step 2, according to reference rotation velocity step two under load torque step under constant reference revolving speed and constant load torque
Kind mode will press the case where load torque variation and under different loads torque the case where reference rotation velocity variation under different reference rotation velocities
It is got according to certain step-length and interval traversal, and the data of the above-mentioned input quantity accordingly generated is sent into surface permanent magnetic and are synchronized
In motor model predictive control algorithm, the optimal voltage vector of Model Predictive Control Algorithm selection in the following control period is generated
Sequence, and by its training sample matrix for being combined into BP neural network corresponding with the value of corresponding each input quantity.
In step 3, the feedforward neural network that BP neural network model is 3 layers, according to hidden layer node empirical equationWherein a is the constant between 0~10, further determines hidden layer section according to commissioning experience on this basis
Point number, input layer correspond to 4 input quantities of above-mentioned determination, and output node layer is equal to the number of alternative voltage vector, BP
The activation primitive of neuron in the hidden layer and output layer of neural network is all made of tansig type tanh transmission function,
The expression formula of tansig function is f (x)=2/ (1+exp (- 2*x)) -1.
In step 4, in order to cancel order of magnitude difference between each dimension data, avoid because of inputoutput data order of magnitude difference
It is larger and cause neural network forecast error larger, it needs that data are normalized, returns in neural network described in this method
One changes processing using traditional default behavior, and processing mode is as follows:
In formula, xmin is the minimum number in data series, and xmax is the maximum number in sequence.Ymin and ymax is specified
Range is normalized, is defaulted as -1 and+1.
In step 5, the process that BP neural network topological model is trained study is as follows:
(5.1) input data:
After carrying out data prediction i.e. data normalization processing to the training data being collected into step 2, one is formed
4 dimension matrixes will be defeated per the one-dimensional dependent variable i.e. input quantity respectively corresponded in surface permanent magnetic Synchronous Machine Models prediction algorithm
Enter and be combined into training sample sequence (X, Y) with output, X is input data, and Y is corresponding output data.
(5.2) netinit:
Network input layer number of nodes n, node in hidden layer l, output layer section are determined according to system input and output sequence (X, Y)
Count m, the connection weight ω between initialization input layer, hidden layer and output layer neuronij, ωjk, initialize hidden layer threshold
Value a, output layer threshold value b give learning rate and neuron excitation function.
(5.3) hidden layer output calculates:
According to input variable X, input layer and implicit interlayer connection weight ωijAnd hidden layer threshold value a, it is defeated to calculate hidden layer
H out.
In formula, l is node in hidden layer, and f is general hidden layer excitation function, selection that there are many functions, letter selected by the present invention
Number is f (x)=2/ (1+exp (- 2*x)) -1
(5.4) output layer output calculates:
H, connection weight ω are exported according to hidden layerjkWith threshold value b, BP neural network prediction output O is calculated.
(5.5) error calculation:
O and desired output Y is exported according to neural network forecast, calculates neural network forecast error J:
ek=Yk-OkK=1,2 ..., m
(5.6) weight threshold updates:
By output layer, according to J, according to gradient descent method direction calculating, and weight threshold is successively adjusted.The step-length is taken to be
η can obtain t+1 adjustment formula of the connection weight and threshold value between neuron u to neuron v:
(5.7) judge whether algorithm iteration terminates, iteration terminates one of the following conditions to be met:
(1) reach the number of iterations 2000；
(2) error predicted drops within target error value 0.05；
(3) continuous 6 verifyings collection error does not reduce or increases instead.If iteration is not over, return step (5.3).
Embodiment:
Surface permanent magnetic Synchronous Machine Models prediction simulation of direct torque control mould is established based on MATLAB/Simulink
Type.Simulation model is discrete model, and the sampling period is 5 × 10-5s.DC bus-bar voltage is 312V.Revolving speed pi regulator parameter
Are as follows: Kp=5, KI=10, pi regulator output are limited to [- 35,35] up and down.It is 0.3Wb with reference to stator magnetic linkage amplitude.Emulation is used
Surface permanent magnetic parameter of synchronous machine is as shown in table 1.
The emulation surface permanent magnetic parameter of synchronous machine of table 1
The study for comprehensively considering two kinds of situations of change of reference rotation velocity and load torque builds training number using ramp function
According to:
In the case of three kinds of reference rotation velocity 10rpm, 30rpm and 60rpm, load torque is arranged using ramp function, just
Initial value is 10Nm, and the end simulation time 2s, 2s load torque rises to 34Nm.
In the case of six kinds of load torque 10Nm, 15Nm, 20Nm, 25Nm, 30Nm and 34Nm, with reference to turn
Speed is arranged using ramp function, and initial value 10rpm, the end simulation time 2s, 2s reference rotation velocity rises to 60rpm.
Above data is collected, by debugging and training, the topological structure of neural network is 4-18-7, and categorizing selection is correct
Rate can reach 88.34%.Neural network is embedded into surface permanent magnetic Synchronous Motor Direct Torque Control System and is gone here and there with MPC
Row design, composition BP-MPC double working modes carry out the work of optimal voltage vector selection, and switchover policy threshold value is arranged after tested
For 2Nm.
Reference rotation velocity is 60rpm, and step is to 30rpm when 1s.Load torque is initially 10N.m, and step is extremely when 0.5s
30N.m.The simulation result under table mode of operation, BP-MPC double working modes and original MPC operating mode is switched respectively as schemed
Shown in 4-Figure 18.
Comparison between different performances is evaluated using following index: the prediction of surface permanent magnetic Synchronous Machine Models is straight
Moment controlling system stable state evaluation index is connect using torque pulsation root-mean-square error Trip_RMSE, magnetic linkage pulse root-mean-square error
ψrip_RMSEIt with average on-off times, is shown below respectively, wherein n is number of samples, NswitchingFor on-off times, t is imitative
The true time.
The evaluation index of 2. 3 kinds of modes of table compares
Emulation experiment shows that motor operation is good under the designs of BP-MPC double working modes, and performance and MPC phase
When even it is slightly good, such as torque pulsation and on-off times, torque pulsation can averagely reduce 1.11%, and maximum can be reduced
8.14%, average switch number can reduce 19.89%.Due to being switched to MPC operating mode when step, dynamic torque is quick
Response performance will not incur loss.Meanwhile the distributed parallel operational capability of neural network alleviates the computational burden of system,
And having the time more than 99% in the entire simulation process of BP-MPC double mode is that BP neural network is participating in system closed loop
Work, MPC are only involved in a small amount of moment.This shows the model prediction algorithm that complexity can be replaced with simple BP neural network, together
When have many advantages, such as that structure is simple, precision is higher, reaction speed is fast, system-computed burden can be mitigated, improve system response and
Shi Xing, and explore application value of the neural network in motor model PREDICTIVE CONTROL.
Way as selection voltage vector temporarily is gone using neural network alternative model prediction algorithm without discovery at present,
So the present invention has initiative, and neural network has the classification capacity of powerful nonlinear fitting ability and pattern-recognition,
The characteristics of its distributed parallel operation, accelerates operation, makes it possible a large amount of operations, goes substitution to calculate using neural network
The bad Model Predictive Control of larger real-time is measured with important exploration meaning and practical application value.
Claims (6)
1. the surface permanent magnetic Synchronous Machine Models forecast Control Algorithm based on BP neural network, which is characterized in that including following
Step:
Step 1 determines input quantity and output quantity in surface permanent magnetic Synchronous Machine Models predictive control algorithm, and nerve
The input quantity and output quantity of network, and determine the variation range of motor reference rotation velocity and load torque；
Step 2, according to reference rotation velocity step two ways under load torque step under constant reference revolving speed and constant load torque
The case where reference rotation velocity changes under the case where load torque under different reference rotation velocities is changed and different loads torque is according to certain
Step-length and interval traversal get, and by the data of the above-mentioned input quantity accordingly generated be sent into surface permanent magnetic Synchronous Machine Models
In predictive control algorithm, the optimal voltage vector sequence of Model Predictive Control Algorithm selection in the following control period is generated, and
By the value of each input quantity training sample for being combined into BP neural network corresponding with the optimal voltage vector that corresponding selection comes out；
Step 3 constructs BP neural network topological model；
Training sample data in step 2 are normalized step 4, and the training sample after normalized is defeated
Enter to have constructed progress off-line training study in BP neural network topological model, obtains satisfactory weight threshold；
The good BP neural network of off-line training is embedded into surface permanent magnetic Synchronous Machine Models prediction Direct torque by step 5
With model prediction algorithm serial design in system processed, motor in the following control period is carried out for alternative model predictive control algorithm
The work and function of optimal voltage vector selection；
A switching value is arranged between model prediction algoritic module and the serial design of BP neural network, works as torque for step 6
When error is greater than certain threshold value, operating mode is switched to the selection work that model prediction algorithm carries out optimal voltage vector by switching value
Make；When torque error value is less than the threshold value, the selection work of voltage vector is carried out using neural network.
2. the surface permanent magnetic Synchronous Machine Models forecast Control Algorithm according to claim 1 based on BP neural network,
It is characterized in that, step 1, determine input quantity in surface permanent magnetic Synchronous Machine Models predictive control algorithm and output quantity with
And the specific method is as follows for neural network input quantity and output quantity:
Ignore rotor rotational movement and Stator resistance voltage dropping, after applying nonzero voltage space vector, subsequent time stator magnetic linkage amplitude and
Torque is as shown in formula 1.1:
It is the amplitude of current K moment stator magnetic linkage,It is the stator magnetic linkage amplitude at k+1 moment,It is currently to want
The voltage vector magnitude of application, fixed amplitude set tentering value for voltage vector, amplitude is constantly equal toIts
Middle UdcIt is DC bus-bar voltage；Δ t is the action time that the voltage vector applies, α be voltage vector and stator magnetic linkage vector it
Between angle；
The torque of subsequent time stator magnetic linkage is as shown in formula 1.2:
Te(k+1) be the k+1 moment motor torque, p is the number of pole-pairs of motor, ψfIt is permanent magnet flux linkage,It is the current K moment
The amplitude of stator magnetic linkage, LdFor d axis stator inductance, α is the angle between voltage vector and stator magnetic linkage vector, and δ (k) is current
The angle of torsion at K moment；
Cost function used is as shown in formula 1.3:
The result that formula 1.1 and formula 1.2 are calculated is sent into cost function, in entire algorithmic procedure, if there is n alternative voltages
Vector can then obtain n cost function numerical value, select voltage vector corresponding to the smallest numerical value of cost function as the calculating
The optimal voltage vector of motor is applied in the control period；
The input quantity of Model Predictive Control Algorithm have respectively angle of torsion, current stator magnetic linkage amplitude, current stator magnetic linkage Angle Position,
Reference Stator Flux Linkage, torque reference, alternative voltage vector angle, output quantity are selected voltage vectors；
Remove constant from input quantity, characteristic dimension is simplified, as the input quantity of neural network, selected voltage vector
As the output quantity of neural network, constant includes Reference Stator Flux Linkage and alternative voltage vector angle.
3. the surface permanent magnetic Synchronous Machine Models forecast Control Algorithm according to claim 1 based on BP neural network,
It is characterized in that, in step 2, it will be in corresponding value range according to certain step-length and interval by programming or the Realization of Simulation
Two kinds of situations of change traversal get.
4. the surface permanent magnetic Synchronous Machine Models forecast Control Algorithm according to claim 1 based on BP neural network,
It is characterized in that, in step 3, the feedforward neural network that BP neural network topological model is three layers, according to hidden layer node experience
FormulaDetermine node in hidden layer, wherein a is the constant between 0~10, is determined according to commissioning experience implicit
Node layer number, input layer correspond to input quantity, and output node layer is equal to the number of alternative voltage vector, and BP neural network is opened up
The activation primitive for the neuron flutterred in the hidden layer and output layer of model is all made of the tanh transmission function of tansig type,
The expression formula of tansig function is f (x)=2/ (1+exp (- 2*x)) -1.
5. the surface permanent magnetic Synchronous Machine Models forecast Control Algorithm according to claim 1 based on BP neural network,
It is characterized in that, normalization uses default treatment mode in step 4, processing mode is as follows:
In formula, xmin is the minimum number in data series, and xmax is the maximum number in sequence, and ymin and ymax are specified normalizing
Change range.
6. the surface permanent magnetic Synchronous Machine Models forecast Control Algorithm according to claim 1 based on BP neural network,
It is characterized in that, BP neural network topological model carries out off-line training study, and the specific method is as follows in step 4:
The first step, the training sample after normalized respectively correspond surface permanent magnetic Synchronous Machine Models prediction algorithm per one-dimensional
In dependent variable, that is, input quantity, input is combined into training sample sequence (X, Y) with output, X is input data, and Y is corresponding
Output data；
Second step, the training sample sequence (X, Y) being combined into according to system input and output determines network input layer number of nodes n, hidden
The l of number containing node layer, output layer number of nodes m, the connection weight ω between initialization input layer, hidden layer and output layer neuronij,
ωjk, hidden layer threshold value a, output layer threshold value b are initialized, learning rate and neuron excitation function are given；
Third step, according to input variable X, input layer and implicit interlayer connection weight ωijAnd hidden layer threshold value a, it calculates implicit
Layer output H；
In formula, l is node in hidden layer, and f is general hidden layer excitation function, selected function are as follows:
F (x)=2/ (1+exp (- 2*x)) -1 1.6
4th step exports H, connection weight ω according to hidden layerjkWith threshold value b, BP neural network prediction output O is calculated；
5th step exports O and desired output Y according to neural network forecast, calculates neural network forecast error J；
ek=Yk-Ok 1.8
Wherein, k=1,2 ..., m；
6th step, according to J, according to gradient descent method direction calculating, and successively adjusts weight threshold, takes step by output layer
A length of η can obtain t+1 adjustment formula of the connection weight and threshold value between neuron u to neuron v:
7th step, judges whether algorithm iteration terminates, if iteration terminates one of the following conditions to be met, iteration terminates:
Condition 1 reaches the number of iterations 2000；
Condition 2, the error of prediction drop within target error value 0.05；
Condition 3, continuous 6 verifyings collection error do not reduce or increase instead；
If iteration is not over, third step is returned.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910817795.8A CN110535396B (en) | 2019-08-30 | 2019-08-30 | Surface permanent magnet synchronous motor model prediction control method based on BP neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910817795.8A CN110535396B (en) | 2019-08-30 | 2019-08-30 | Surface permanent magnet synchronous motor model prediction control method based on BP neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110535396A true CN110535396A (en) | 2019-12-03 |
| CN110535396B CN110535396B (en) | 2021-03-30 |
Family
ID=68665648
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910817795.8A Expired - Fee Related CN110535396B (en) | 2019-08-30 | 2019-08-30 | Surface permanent magnet synchronous motor model prediction control method based on BP neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110535396B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112691783A (en) * | 2020-12-09 | 2021-04-23 | 华润电力技术研究院有限公司 | Pulverized coal boiler pulverizing system regulation and control method, device, equipment and storage medium |
| CN115085627A (en) * | 2022-08-22 | 2022-09-20 | 成都微精电机股份公司 | Motor parameter dynamic identification method |
| CN117277888A (en) * | 2023-11-23 | 2023-12-22 | 潍柴动力股份有限公司 | Method and device for predicting voltage vector of stator of permanent magnet synchronous motor |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102335872A (en) * | 2011-09-14 | 2012-02-01 | 桂林电子科技大学 | Artificial neural network-based method and device for automatically trimming grinding wheel of grinding machine |
| CN105262358A (en) * | 2015-11-11 | 2016-01-20 | 国网上海市电力公司 | Bidirectional grid connected inverter |
| JP2017211913A (en) * | 2016-05-27 | 2017-11-30 | 日本電信電話株式会社 | Behavior determination device, future prediction model learning device, network learning device, method, and program |
| CN108880394A (en) * | 2018-06-04 | 2018-11-23 | 江苏大学 | A kind of wavelet neural network position-less sensor of switched reluctance motor forecast Control Algorithm |
| CN109760523A (en) * | 2019-01-21 | 2019-05-17 | 浙江工业大学 | Composite power source energy management method based on BP neural network speed prediction |
| CN109873581A (en) * | 2017-12-04 | 2019-06-11 | 北京自动化控制设备研究所 | A kind of direct method for control speed of permanent magnet synchronous motor neural network based |
-
2019
- 2019-08-30 CN CN201910817795.8A patent/CN110535396B/en not_active Expired - Fee Related
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102335872A (en) * | 2011-09-14 | 2012-02-01 | 桂林电子科技大学 | Artificial neural network-based method and device for automatically trimming grinding wheel of grinding machine |
| CN105262358A (en) * | 2015-11-11 | 2016-01-20 | 国网上海市电力公司 | Bidirectional grid connected inverter |
| JP2017211913A (en) * | 2016-05-27 | 2017-11-30 | 日本電信電話株式会社 | Behavior determination device, future prediction model learning device, network learning device, method, and program |
| CN109873581A (en) * | 2017-12-04 | 2019-06-11 | 北京自动化控制设备研究所 | A kind of direct method for control speed of permanent magnet synchronous motor neural network based |
| CN108880394A (en) * | 2018-06-04 | 2018-11-23 | 江苏大学 | A kind of wavelet neural network position-less sensor of switched reluctance motor forecast Control Algorithm |
| CN109760523A (en) * | 2019-01-21 | 2019-05-17 | 浙江工业大学 | Composite power source energy management method based on BP neural network speed prediction |
Non-Patent Citations (1)
| Title |
|---|
| BEN GUO ETAL.: "NN-based model predictive direct speed control of PMSM drive systems", 《2014 INTERNATIONAL CONFERENCE ON MACHINE LEARNING AND CYBERNETICS》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112691783A (en) * | 2020-12-09 | 2021-04-23 | 华润电力技术研究院有限公司 | Pulverized coal boiler pulverizing system regulation and control method, device, equipment and storage medium |
| CN115085627A (en) * | 2022-08-22 | 2022-09-20 | 成都微精电机股份公司 | Motor parameter dynamic identification method |
| CN117277888A (en) * | 2023-11-23 | 2023-12-22 | 潍柴动力股份有限公司 | Method and device for predicting voltage vector of stator of permanent magnet synchronous motor |
| CN117277888B (en) * | 2023-11-23 | 2024-03-19 | 潍柴动力股份有限公司 | Method and device for predicting voltage vector of stator of permanent magnet synchronous motor |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110535396B (en) | 2021-03-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110535146B (en) | Electric power system reactive power optimization method based on depth determination strategy gradient reinforcement learning | |
| CN110266228A (en) | Surface permanent magnetic Synchronous Machine Models forecast Control Algorithm based on BP neural network | |
| Le et al. | Self-evolving type-2 fuzzy brain emotional learning control design for chaotic systems using PSO | |
| CN110535396A (en) | Surface permanent magnetic Synchronous Machine Models forecast Control Algorithm based on BP neural network | |
| CN104776446B (en) | Combustion optimization control method for boiler | |
| CN109946975A (en) | A kind of intensified learning optimal track control method of unknown servo-system | |
| CN109327178A (en) | A kind of switched reluctance machines Multi-step predictive control device building method | |
| CN110518847A (en) | Surface permanent magnetic Synchronous Machine Models forecast Control Algorithm based on BP neural network | |
| CN111608868B (en) | Maximum power tracking adaptive robust control system and method for wind power generation system | |
| CN111812968B (en) | Fuzzy neural network PID controller-based valve position cascade control method | |
| CN109254530A (en) | MFA control method based on grinding process basis circuit | |
| CN110518860A (en) | Permanent magnet synchronous motor model predictive control method based on BP neural network and switch list | |
| Zhao et al. | Self-Tuning of PID parameters based on adaptive genetic algorithm | |
| CN102393645A (en) | Control method of high-speed electro-hydraulic proportional governing system | |
| Zhao | Adaptive Fuzzy Control of a Class of Discrete-Time Nonlinear Systems | |
| CN110083055A (en) | A kind of PID automatic setting method based on deep learning and LOGFA | |
| CN109932909A (en) | The big system of fired power generating unit desulphurization system couples Multi-variables optimum design match control method | |
| CN106371321A (en) | PID control method for fuzzy network optimization of coking-furnace hearth pressure system | |
| CN110888323A (en) | Control method for intelligent optimization of switching system | |
| Menghal et al. | Application of artificial intelligence controller for dynamic simulation of induction motor drives | |
| Guolian et al. | Multiple-model predictive control based on fuzzy adaptive weights and its application to main-steam temperature in power plant | |
| Kaminski et al. | Neural speed controller based on two state variables applied for a drive with elastic connection | |
| Neto et al. | An improved fuzzy learning algorithm for motion control applications [PM synchronous motors] | |
| Chao et al. | Design of reinforcement learning algorathm for single inverted pendulum swing control | |
| Zhu et al. | Memristor-based neural network PID controller for buck converter |
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 | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20210330 Termination date: 20210830 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |