CN102509152A - Switched reluctance motor on-line modeling method based RBF neural network - Google Patents
Switched reluctance motor on-line modeling method based RBF neural network Download PDFInfo
- Publication number
- CN102509152A CN102509152A CN2011103486658A CN201110348665A CN102509152A CN 102509152 A CN102509152 A CN 102509152A CN 2011103486658 A CN2011103486658 A CN 2011103486658A CN 201110348665 A CN201110348665 A CN 201110348665A CN 102509152 A CN102509152 A CN 102509152A
- Authority
- CN
- China
- Prior art keywords
- switched reluctance
- neural network
- line
- reluctance motor
- 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.)
- Pending
Links
Images
Landscapes
- Feedback Control In General (AREA)
- Control Of Electric Motors In General (AREA)
Abstract
The invention discloses a switched reluctance motor on-line modeling method based a RBF (Radial Basis Function) neural network, belonging to the technical field of intelligent control on the switched reluctance motor. The method is used for establishing an off-line model of the switched reluctance motor through a RBF neural network method based on the static data of the switched reluctance motor. On that basis, an error regulation method is designed, the on-line model of the switched reluctance motor is established, and the on-line model has the real-time on-line regulation function and can describe the dynamic characteristics of the switched reluctance motor more accurately. In an experimental process, a method of establishing an input-output mapping relation is used in the on-line modeling method, the problem of overlong operation time of the Gaussian function in a DSP (Digital Signal Processor) is solved, and the realizability of a simulation model in engineering application is ensured, wherein the simulation model is designed based the RBF neural network.
Description
Technical field
The present invention relates to a kind of switched reluctance machines line modeling method, belong to the switched reluctance machines field of intelligent control based on the RBF neural network.
Background technology
Because SRM gets the biconvex electrode structure and the magnetic circuit height is saturated, so, be difficult to obtain the analytic expression of magnetic linkage ψ, so be difficult to set up the SRM precise math model as the suitable complicacy of the calculating of the magnetization curve of the basis-magnetic linkage-electric current-rotor-position of SRM various characteristics.The classic method of SRM being carried out modeling has: linear approach, function analytical method, and limited element analysis technique, but for above-mentioned reasons, the model that uses these traditional methods to set up can't satisfy the accuracy requirement that complicated control method proposes gradually.
Along with the development of intelligence control method is used with ripe; Utilize intelligence control method more and more to the achievement of SRM modeling; The degree of accuracy of the model of setting up through intelligent method is also increasingly high; Through years of researches, numerous scholars successfully with fuzzy control, genetic algorithm and Application of Neural Network in the modeling of SRM.Mainly comprised by the frequent neural network of using: B SPLINE NEURAL NETWORK (BSNN), artificial neural network (ANN), reverse transmittance nerve network (BPNN), radial basis function neural network (RBFNN) etc.From the document of having delivered, can sum up following some deficiency: 1, department pattern reckons without the situation that possibly run in the motor practical engineering application, and they can only be applicable to simulated environment; Though 2, department pattern is verified on motor working control platform, the static measurement data that this department pattern all is based on motor are that off-line data is set up, and therefore can not describe the dynamic perfromance of motor exactly; 3, department pattern has had the dynamic adjustments function, and on motor working control platform, verifies, but this department pattern the priori of emulation or the priori conditions of experiment are had relatively high expectations, the religion of the usable range of algorithm is narrow, portability is not strong.
Summary of the invention
The problem that the present invention will solve is to realize respectively switched reluctance machines being set up accurate off-line model and at line model, and guarantees the engineering realizability of two kinds of models.To this problem, the present invention proposes a kind of modeling method based on the RBF neural network, thus the accuracy of the nonlinear model that has guaranteed to be set up; The present invention also carried out simplifying to basis function radially and handled, thus solved neural network in DSP working time process and limit the problem of its widespread use in Project Realization
The present invention adopts following technical scheme for realizing above-mentioned purpose:
The present invention proposes a kind of method of new switched reluctance machines modeling, comprise the steps: on the basis of SRM static measurement data, training RBF neural network module, and set up the off-line model of SRM based on this; On the basis of off-line model, add Error Feedback, the design error control method, set up SRM at line model; In process of the test, done the simplification processing to moving the longest RBF consuming time, thereby realized the long practical application of neural network realistic model in engineering consuming time.
This method of SRM being carried out line modeling based on the RBF neural network that the present invention adopts; Mainly contain a little: the present invention has adopted a kind of new method to handle Gaussian function; Make calculating shared time in the dsp operation process of Gaussian function reduce significantly, and guaranteed computational accuracy; The present invention has carried out experimental verification with the simulation algorithm of design, has guaranteed the feasibility of simulation algorithm in practical engineering application; The present invention has added the error control method, makes model have the dynamic adjustments function, can describe the dynamic perfromance of SRM more exactly; The simulation algorithm that the present invention designed (comprising off-line model algorithm and online model algorithm) coupled mode is good, and is portable strong, can be widely used in other field.
Description of drawings
Fig. 1 is switched reluctance machines model control synoptic diagram.The PID controller is a proportional-integral derivative controller, and the APC controller is angle position control, and the CCC controller is current chopping control.
The SRM realistic model that Fig. 2 is set up down for SIMULINK environment among the MATLAB.
Fig. 3 is the static φ _ θ _ i relation curve synoptic diagram of 7.5KW switched reluctance machines, and φ is a magnetic linkage, and θ is an angle, and i is an electric current.
The RBF neural network module inner structure of Fig. 4 for training among the MATLAB.
Fig. 5 is the inner structure of Layer1 module among Fig. 4.
Fig. 6 is the inner structure of Layer2 module among Fig. 4.
Fig. 7 is the RBF neural metwork training process error synoptic diagram that successively decreases.
Fig. 8 is a RBF neural network test result synoptic diagram.
Fig. 9 is the input and output mapping relations synoptic diagram of Gaussian function, and x is the input codomain, and y is the output codomain.
Off-line model output magnetic linkage and magnetic linkage Error Simulation synoptic diagram when Figure 10 is 2000r/min for rotating speed.
Off-line model output magnetic linkage experiment synoptic diagram when Figure 11 is 2000r/min for rotating speed.
Figure 12 regulates output weights synoptic diagram for the RBF neural network dynamic, is depicted as RBF structural representation, ω in the square frame of the left side
NewBe the up-to-date weights after regulating, ω
OldBe the weights before regulating, η is a learning rate, e
ψFor estimating that magnetic linkage and expectation magnetic linkage are the error between the actual magnetic linkage of motor.
Online model emulation synoptic diagram as a result when Figure 13 is 2000r/min for rotating speed.
Online model experiment synoptic diagram as a result when Figure 14 is 2000r/min for rotating speed.
Figure 15 undergos mutation for the expectation magnetic linkage and estimates the dynamic adjustments process synoptic diagram of magnetic linkage under the situation.Cyan is the expectation magnetic linkage among the figure, and is red for estimating magnetic linkage.A expects that constantly magnetic linkage undergos mutation, and estimates that magnetic linkage begins dynamic adjustments, and B expects magnetic linkage on the estimated magnetic flux D-chain trace constantly.
Embodiment
Merit the present invention provides a kind of method of SRM being carried out modeling based on the RBF neural network; Mainly comprise two parts: (one) off-line modeling: the flux linkage characteristic curve that obtains SRM through the two-dimensional finite meta analysis; And utilize this data set training RBF neural network, realize the off-line modeling of SRM magnetic linkage; (2) the online adjusting of model: according to the variation of running status; The estimation magnetic linkage and the actual detected magnetic linkage of off-line model can produce deviation; Output weights through to the RBF neural network carry out online adjusting, and the SRM that foundation has the line dynamic regulatory function is at line model.
Be elaborated below in conjunction with the technical scheme of accompanying drawing to invention:
As shown in Figure 1; Conventional SRM control method has angle position control and current chopping control; The method that employing of the present invention combines is carried out modeling to SRM; Adopt current chopping control during slow running, adopt angle position control during the high speed operation, in the SIMULINK environment, build the realistic model (as shown in Figure 2) of SRM based on this.
Two characteristics of SRM most critical are magnetic linkage-phase current-rotor-position characteristic and torque-phase current-rotor-position characteristic; Therefore set up an accurate switched reluctance machines model, the quality of required training sample directly has influence on the accuracy and the generalization ability of the model of setting up.Obviously, when training sample distributes rationally, cover when comprehensive, the model of being set up has higher relatively precision and generalization ability preferably.Because identical with the principle that torque characteristics carries out modeling to flux linkage characteristic based on the RBF neural network, the present invention only studies flux linkage characteristic, result of study can be applied in the modeling of torque characteristics equally.The flux linkage characteristic Data Acquisition generally has 2 kinds of methods: measurement method and limited element analysis technique, the employed data of this paper obtain through limit meta analysis method.To 0 °≤θ of rotor-position≤22.5 °, the flux linkage characteristic in electric current 0A≤i≤12.5A scope carries out finite element analysis, and obtains data (as shown in Figure 3).
The data set (as shown in Figure 3) that utilizes finite element analysis to obtain, training RBF neural network module in MATLAB, available RBF mixed-media network modules mixed-media inner structure (like Fig. 4) in the MATLAB simulated environment.
Fig. 5 is the inner structure of Layer1 module among Fig. 4.
Fig. 6 is the inner structure of Layer2 module among Fig. 4.
In the MATLAB environment, set the anticipation error of training, when actual error met the expectation error requirements, training process finished (as shown in Figure 7).
The RBF neural metwork training is imported some groups of input signals after finishing at random, and the mixed-media network modules mixed-media of being trained is detected, and testing result (as shown in Figure 8) shows that the RBF mixed-media network modules mixed-media precision of training meets the demands.
In the practical applications of RBF neural network, the Gaussian function holding time is long, and to cause program incorrect at dsp operation be a great problem.The present invention is directed to this difficult problem and proposed a kind of new solution, can find out through the input and output mapping characteristics (as shown in Figure 9) of observing Gaussian function, have only input value when Gaussian function within certain scope, its result of calculation just can be identified.Therefore, set up the input and output mapping table in [3.5 ,+3.5] scope, investigate and prosecute output accordingly in this scope through the method for tabling look-up, not in this scope if input can be set and be output as 0.Use this method, the computing time of Gaussian function and computational accuracy all obviously are superior to the Taylor expansion method.
Based on said method, set up the off-line simulation model of SRM, can observe the output magnetic linkage of SRM and the error (shown in figure 10) between the actual magnetic linkage under the MATLAB environment.
According to the method for above-mentioned simplification processing RBF, the DSP program of writing the off-line model algorithm, control SRM operation, and output of estimation magnetic linkage and the actual magnetic linkage of observation SRM are exported (shown in figure 11).
Above model of setting up and the employed method of experiment all are to be based upon on the static measurement data of SRM, therefore the dynamic perfromance of accurate description motor all sidedly.
In the process of motor actual motion, tend to take place motor operating state and (comprise rotating speed, torque; Voltage fluctuation; Situation about external interference, environmental factor or the like) changing, under these circumstances; Based on the off-line model that off-line data is set up, the real-time magnetic linkage-electric current of motor-rotor-position relation can't be described exactly.
In order to address this problem, the model of being set up is operated under the various running status at SRM, the SRM flux linkage characteristic can be described exactly in real time; This paper has designed the online control method of RBF network output weights; When the expectation running status of motor changes, produce error between the magnetic linkage meeting that off-line model is estimated and the actual magnetic linkage of motor, when error during greater than setting threshold; Reduce the output weights of RBF network, estimate that magnetic linkage can reduce thereupon; When error during less than setting threshold, increase the output weights of RBF network, estimate that magnetic linkage can increase thereupon; When error between setting threshold, think that then this error constantly satisfies estimation accuracy, need not weights are regulated.Through above-mentioned adjustment process, the estimation magnetic linkage can be followed the tracks of actual magnetic linkage in a short period of time, has guaranteed the estimated accuracy at line model.(shown in figure 12), set up SRM at line model, can be according to the error between RBF network magnetic linkage estimated result and the actual measurement magnetic linkage result, online in real time study and from main regulation makes the RBF network output magnetic linkage actual magnetic linkage of tracking motor better.As can be seen from Figure 12, the error control method of the present invention's employing is gradient descent method: ω
New=ω
Old+ η * e
ψThe specification error threshold value is ± 1E-3Wb; If the magnetic linkage error exceeds the error threshold scope; The output weights of RBF neural network can be with carrying out dynamic adjustments according to regulating algorithm on-line; Through emulation experiment repeatedly, choose suitable learning rate η value, can improve constantly the degree of accuracy of magnetic linkage model through on-line study.
According to the method described above, the emulation of setting up SRM is at line model, and observation magnetic linkage output (shown in figure 13).
According to the method for above-mentioned simplification processing RBF, be programmed in the DSP program of line model algorithm, the operation of control SRM, and detect actual magnetic chain result and estimated magnetic flux chain result (shown in figure 14).
Have the dynamic adjustments function in order further to verify at line model, the present invention has designed the experiment (shown in figure 15) that the expectation magnetic linkage is undergone mutation.Figure 15 shows, A expects that constantly magnetic linkage undergos mutation, and the RBF neural network module can onlinely carry out dynamic adjustments, progressively dwindles the error between the two, finally makes constantly that at B estimating that magnetic linkage can be followed the tracks of expects magnetic linkage.
Claims (2)
1. the switched reluctance machines based on the RBF neural net method carries out the line modeling method, it is characterized in that comprising the steps:
1) utilize the static data of switched reluctance machines to train the RBF neural network module;
2) on the basis of off-line model, added the error control method, set up SRM at line model;
3) Gaussian function has been set up the input-output function mapping table;
4) the present invention is converted into executable code with the off-line model of SRM with at line model on the basis of step 3), and with its control SRM operation.
2. a kind of switched reluctance machines based on the RBF neural net method according to claim 1 carries out the line modeling method, it is characterized in that said input-output function mapping table is in [3.5 ,+3.5] scope.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2011103486658A CN102509152A (en) | 2011-11-08 | 2011-11-08 | Switched reluctance motor on-line modeling method based RBF neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2011103486658A CN102509152A (en) | 2011-11-08 | 2011-11-08 | Switched reluctance motor on-line modeling method based RBF neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN102509152A true CN102509152A (en) | 2012-06-20 |
Family
ID=46221232
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN2011103486658A Pending CN102509152A (en) | 2011-11-08 | 2011-11-08 | Switched reluctance motor on-line modeling method based RBF neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN102509152A (en) |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103095191A (en) * | 2013-01-29 | 2013-05-08 | 中国矿业大学 | Switch reluctance motor memory sensor model modeling method |
| CN103177155A (en) * | 2013-02-28 | 2013-06-26 | 重庆科技学院 | Oilfield pumping unit oil pumping energy saving and production increasing optimization method based on back propagation neural network (BPNN) and strength Pareto evolutionary algorithm 2 (SPEA2) |
| WO2014063452A1 (en) * | 2012-10-22 | 2014-05-01 | 中国矿业大学 | Memristor linear modeling method for switched reluctance motor |
| CN103886131A (en) * | 2014-02-25 | 2014-06-25 | 江苏大学 | Switch reluctance motor magnetic flux linkage online modeling method based on extreme learning machine |
| WO2016184110A1 (en) * | 2015-05-15 | 2016-11-24 | 中国矿业大学 | Switched reluctance motor modeling method |
| CN103886131B (en) * | 2014-02-25 | 2016-11-30 | 江苏大学 | A kind of switched reluctance motor flux linkage line modeling method based on extreme learning machine |
| CN107196565A (en) * | 2017-07-04 | 2017-09-22 | 江苏理工学院 | A kind of Computation of Nonlinear Characteristics on Switched Reluctance Motor line modeling method |
| CN108448952A (en) * | 2018-05-12 | 2018-08-24 | 烟台仙崴机电有限公司 | A kind of energy the Internet switch magnetic resistance motor rotor location estimation method |
| CN108549752A (en) * | 2018-03-27 | 2018-09-18 | 南京航空航天大学 | A kind of Fielding-winding doubly salient generator functional level model modelling approach |
| CN110543682A (en) * | 2019-08-03 | 2019-12-06 | 湖南贝加尔动力科技有限公司 | SRM opening angle optimization method based on low inductance region nonlinear inductance model |
| CN112928965A (en) * | 2021-03-29 | 2021-06-08 | 桂林电子科技大学 | Flux linkage based torque ripple suppression control system and method for switched reluctance motor |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1655438A (en) * | 2005-03-11 | 2005-08-17 | 江苏大学 | Magnetic levitation switch reluctance motor radial neural network reversed decoupling controller and method for constructing same |
-
2011
- 2011-11-08 CN CN2011103486658A patent/CN102509152A/en active Pending
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1655438A (en) * | 2005-03-11 | 2005-08-17 | 江苏大学 | Magnetic levitation switch reluctance motor radial neural network reversed decoupling controller and method for constructing same |
Non-Patent Citations (3)
| Title |
|---|
| 姚雪莲等: "基于在线模糊神经网络建模的开关磁阻电机高性能转矩控制", 《电机与控制应用》 * |
| 纪良文等: "基于径向基函数神经网络的开关磁阻电机建模", 《电工技术学报》 * |
| 薛梅等: "基于DSP的开关磁阻电机磁链特性检测与神经网络建模", 《电工技术学报》 * |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9715479B2 (en) | 2012-10-22 | 2017-07-25 | China University Of Mining And Technology | Memristor linear modeling method for switched reluctance motor |
| WO2014063452A1 (en) * | 2012-10-22 | 2014-05-01 | 中国矿业大学 | Memristor linear modeling method for switched reluctance motor |
| CN103095191B (en) * | 2013-01-29 | 2014-12-10 | 中国矿业大学 | Switch reluctance motor memory sensor model modeling method |
| CN103095191A (en) * | 2013-01-29 | 2013-05-08 | 中国矿业大学 | Switch reluctance motor memory sensor model modeling method |
| CN103177155A (en) * | 2013-02-28 | 2013-06-26 | 重庆科技学院 | Oilfield pumping unit oil pumping energy saving and production increasing optimization method based on back propagation neural network (BPNN) and strength Pareto evolutionary algorithm 2 (SPEA2) |
| CN103177155B (en) * | 2013-02-28 | 2016-04-20 | 重庆科技学院 | A kind of oil-field oil pumper oil recovery energy-saving and production-increase optimization method based on BP neural network and SPEA2 algorithm |
| CN103886131A (en) * | 2014-02-25 | 2014-06-25 | 江苏大学 | Switch reluctance motor magnetic flux linkage online modeling method based on extreme learning machine |
| CN103886131B (en) * | 2014-02-25 | 2016-11-30 | 江苏大学 | A kind of switched reluctance motor flux linkage line modeling method based on extreme learning machine |
| WO2016184110A1 (en) * | 2015-05-15 | 2016-11-24 | 中国矿业大学 | Switched reluctance motor modeling method |
| CN107196565A (en) * | 2017-07-04 | 2017-09-22 | 江苏理工学院 | A kind of Computation of Nonlinear Characteristics on Switched Reluctance Motor line modeling method |
| CN108549752A (en) * | 2018-03-27 | 2018-09-18 | 南京航空航天大学 | A kind of Fielding-winding doubly salient generator functional level model modelling approach |
| CN108448952A (en) * | 2018-05-12 | 2018-08-24 | 烟台仙崴机电有限公司 | A kind of energy the Internet switch magnetic resistance motor rotor location estimation method |
| CN110543682A (en) * | 2019-08-03 | 2019-12-06 | 湖南贝加尔动力科技有限公司 | SRM opening angle optimization method based on low inductance region nonlinear inductance model |
| CN112928965A (en) * | 2021-03-29 | 2021-06-08 | 桂林电子科技大学 | Flux linkage based torque ripple suppression control system and method for switched reluctance motor |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN102509152A (en) | Switched reluctance motor on-line modeling method based RBF neural network | |
| CN108303885A (en) | A kind of motor position servo system self-adaptation control method based on interference observer | |
| CN102385342B (en) | Self-adaptation dynamic sliding mode controlling method controlled by virtual axis lathe parallel connection mechanism motion | |
| CN107179682A (en) | A kind of passive type load simulator and Surplus Moment suppressing method | |
| CN104533701A (en) | Automatic setting method for control parameters of water turbine speed regulating system | |
| CN103984234A (en) | Electro hydraulic servo system self-correction fuzzy PID control method | |
| CN103197596B (en) | A kind of digital control processing parameters self-adaptive fuzzy control rule optimization method | |
| CN106877746A (en) | Method for control speed and speed control unit | |
| CN109787251B (en) | Cluster temperature control load aggregation model, system parameter identification and reverse control method | |
| CN106873380A (en) | Piezoelectric ceramics fuzzy PID control method based on PI models | |
| CN105888971A (en) | Active load reducing control system and method for large wind turbine blade | |
| CN109885077A (en) | A kind of quadrotor attitude control method and controller | |
| CN110703693A (en) | Iterative learning feedforward control method and system for machine tool feeding system | |
| CN109656140A (en) | A kind of fractional order differential offset-type VSG control method | |
| CN105978400A (en) | Ultrasonic motor control method | |
| CN105573120B (en) | Non-linear more single pendulum network system control method for coordinating based on multiple agent | |
| CN107678276A (en) | A kind of adaptive composite control method based on turning table control | |
| CN102269971A (en) | Self-adapting servo controller based on model tracking | |
| CN107728481B (en) | Closed-loop modeling method and device based on model predictive control | |
| CN108988710A (en) | Consider the networking H ∞ model reference DC motor speed-regulating method and system of long delay | |
| CN108710293A (en) | A kind of fault-tolerant iterative learning control method of Direct Current Governor System | |
| Arshad et al. | Deep Deterministic Policy Gradient to Regulate Feedback Control Systems Using Reinforcement Learning. | |
| CN114114929B (en) | Unmanned vehicle path tracking method based on LSSVM | |
| CN102647130A (en) | Permanent magnet synchronous linear motor control method | |
| CN109245665B (en) | Motor servo control method based on data learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C02 | Deemed withdrawal of patent application after publication (patent law 2001) | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20120620 |