CN102937670B - Method for identifying resistance parameters of rotors of induction motor on basis of Elman neural network - Google Patents
Method for identifying resistance parameters of rotors of induction motor on basis of Elman neural network Download PDFInfo
- Publication number
- CN102937670B CN102937670B CN201210463391.1A CN201210463391A CN102937670B CN 102937670 B CN102937670 B CN 102937670B CN 201210463391 A CN201210463391 A CN 201210463391A CN 102937670 B CN102937670 B CN 102937670B
- Authority
- CN
- China
- Prior art keywords
- input
- layer
- output
- network
- error
- 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.)
- Expired - Fee Related
Links
Landscapes
- Control Of Ac Motors In General (AREA)
- Control Of Electric Motors In General (AREA)
Abstract
Disclosed is a method for identifying resistance parameters of rotors of induction motors on the basis of an Elman neural network. The method has the advantages that the Elman neural network and a network structure are determined, a training sample is processed, and the parameters are identified, so that a reference model can be obtained when speed regulating signals are different, and requirements of identification for the resistance parameters of the rotors of the asynchronous motors on rotational speed sensors are met.
Description
Technical field
The invention belongs to and belong to asynchronous machine field, relate to a kind of high performance frequency conversion speed-adjusting system parameter identification method.
Background technology
When motor running, due to the impact of internal and external condition, the parameter of itself can change.The change of electric machine temperature rise and frequency all can have influence on rotor resistance, and it about has 50% with motor temperature change is the highest, and when rotor current frequency is higher, the rotor resistance change that kelvin effect causes can reach several times.Its change can cause the change of rotor time constant etc., causes the various motor feedback signals distortions of calculating based on preset parameter setting.Based on such feedback, the directed coordinate of motor-field often departs from reality, and cause larger rotating speed, torque departs from or pulse, control system performance can be had a greatly reduced quality.Therefore, in the middle of frequency conversion speed-adjusting system runs, need constantly to adjust the parameter of electric machine in each computation model, change to make it follow true parameter of electric machine value, thus guarantee correct close-loop feedback, the performance of Guarantee control system.
Summary of the invention
The present invention is for solving the problems of the technologies described above, based on the MRAS scheme of flux linkage model, propose a kind of induction electromotor rotor resistance parameter discrimination method based on Elman neural network, to solve when speed conditioning signal different, the acquisition problem of reference model, and to the requirement to speed probe during asynchronous electromotor rotor resistance parameter identification.
The present invention is the technical scheme that the deficiency solved the problems of the technologies described above adopts: a kind of induction electromotor rotor resistance parameter discrimination method based on Elman neural network, comprises following step:
Step one, determine Elman neural network: the expression formula of Elman Neural Network Based Nonlinear state space is
;
;
, in formula: k, m, n represent input layer, hidden layer, the neuronic number of output layer respectively, P, T represent input and output vector,
represent the individual weights be input between m hidden neuron of kth,
represent the weights between m hidden neuron to the n-th output layer neuron,
represent m the weights accepted between layer to m hidden neuron,
,
represent the transport function of hidden layer and output layer respectively,
,
the input represented respectively between each layer neuron is biased,
,
represent the feedback states vector of hidden layer output vector and its undertaking layer respectively, adopt a kind of like this algorithm of optimization:
, Elman neural network is optimized, in formula,
modified weight value when being n-th iteration,
for speedup factor,
for factor of momentum;
Step 2, determine network structure: first, specify the input/output argument of Elman neural network, the principal element had an impact to rotor resistance is rotor current, frequency and environment temperature, make a comprehensive input variable to select electric motor winding end temperature; With rotor resistance
as output variable;
The acquisition of step 3, training sample: using the input signal in step 2 as master control condition, combine mutually with other input variable, adopt BP off-line algorithm to carry out training and obtain a collection of training sample, rotor resistance model is set up as reference using these training sample data, with the output of this model as the target of network to value, carry out error-duration model and modified weight;
The process of step 4, training sample: process the training sample that step 3 obtains, when inputting data of different nature, is normalized input data, make it entirely to drop on ± 1 scope in, to facilitate training and the emulation of network;
Step 5, rotor resistance identification: adopt method from simple to complex, first to have the single parameter of input variable identification compared with definite relation, after its identification effect is compared and is verified, add other underlying factor gradually again, compare again, analyze, with the training sample handled well in step 4 for condition, identification is carried out to rotor resistance, first using on its most significant temperature of impact and slip frequency as input, its effect is detected with network, add electric current again as input, whether the identification effect of supervising network improves, finally determine input parameter number and the hidden neuron number of network.
Training sample acquisition methods of the present invention is,
Step one, set up BP network with the input and output parameter in step 3: input vector is
, hidden layer output vector:
, output layer output vector:
, desired output vector:
, the weight matrix between input layer to hidden layer represents with V,
, wherein column vector
for the weight vector that a hidden layer jth neuron is corresponding, the weight matrix between hidden layer to output layer represents with W,
, wherein column vector
for the weight vector that an output layer kth neuron is corresponding.
Step 2, determine error E: for output layer, have
,
, wherein
, for hidden layer, have
,
, wherein
, in above-mentioned function
be unipolarity Sigmoid function,
, when network exports and desired output does not wait, there is output error E, be defined as follows
, above error definition is expanded to hidden layer, has
, be expanded to input layer further, have
;
Step 3, modified weight: from above-mentioned expression formula, network error originated from input is each layer weights
function, adjustment weights change error E, adjustment weights error is constantly reduced, the adjustment amount of weights is directly proportional, namely to the Gradient Descent of error
with
.
The disposal route of training sample of the present invention is, when parameters normalization, relate to the select permeability of maximal value, quantification when this samples with AD is identical; By all inputs simultaneously divided by this quantized value, all inputs be limited to ± 1 scope within, about the quantification of temperature input, can determine according to the class of insulation of motor; Choosing about current maxima, adopts the maximum current of threephase asynchronous machine, and when appearing at starting, maximum starting current is about the 4-7 of its ratings doubly; The maximal value of slip frequency is when appearing at electric motor starting, the maximum slip frequency of power taking machine in operational process.
Beneficial effect of the present invention is: when motor running, and due to the impact of internal and external condition, the parameter of itself can change, the directed coordinate of motor-field often departs from reality, cause larger rotating speed, torque departs from or pulse, control system performance can be had a greatly reduced quality.Therefore, in the middle of frequency conversion speed-adjusting system runs, need constantly to adjust the parameter of electric machine in each computation model, change to make it follow true parameter of electric machine value, thus guarantee correct close-loop feedback, the performance of Guarantee control system.When adopting this method to carry out identification to rotor resistance, first analysis is made to the factor that it has an impact, draw identification input quantity, then it is carried out to the process of training sample, utilize the sample handled well, identification is carried out to rotor resistance, finally draws suitable network structure.Because Elman neural network feedback type neural network has the total characteristic of neural network, can approach arbitrary function by arbitrary accuracy in theory, directly can reflect the characteristic of dynamic process system, be applicable to by the problem solving the identification of rotor resistance parameter.
Accompanying drawing explanation
Accompanying drawing 1 is speed adaptive stator observer;
Accompanying drawing 2 is Elman network structure;
Accompanying drawing 3 is BP neural network;
Accompanying drawing 4 is copper cash electrical resistance rate of temperature change;
Accompanying drawing 5 is that rotor resistance is with slip frequency rate of change;
Accompanying drawing 6 compares with network output valve for Rr test sample book target exports;
Accompanying drawing 7 is that Rr test sample book target exports the graph of errors with network output valve;
Accompanying drawing 8 is Rr test sample book after newly-increased electric current input;
Accompanying drawing 9 is analyzed for network after newly-increased electric current input exports
Embodiment
As shown in the figure, a kind of induction electromotor rotor resistance parameter discrimination method based on Elman neural network, comprises following step:
The first step, determines Elman neural network.Elman neural network is a kind of two-layer BP network structure with feedback, and its feedback link outputs to its input end from hidden layer, and it can store previous value, and is applied in the present invention.Feedback states is different, then Output rusults is different, this feedback system make Elman network can detecting and identifying spatial model and time become pattern.The same with other neural network, the process of the dynamic evolution of each neuron link weight coefficients is depended in its study and identification, and the network trained can produce response for time domain and spatial domain problem, as shown in Figure 2, therefrom can find out, except the input that general networking has
, network weight
, biased
, transport function
, also there is the feedback element from outputting to input
.
The non-linear state space of Elman network can be expressed as:
,
,
, in formula: k, m, n represent input layer, hidden layer, the neuronic number of output layer respectively.P, T represent input and output vector.
,
,
represent the individual weights be input between m hidden neuron of kth, the weights between m hidden neuron to the n-th output layer neuron, m the weights accepted between layer to m hidden neuron.
,
represent the transport function of hidden layer and output layer respectively.
,
the input represented respectively between each layer neuron is biased.
,
represent the feedback states vector of hidden layer output vector and its undertaking layer respectively.
The weighed value adjusting function of Elman network is the same with BP network exists speed of convergence comparatively slowly, and there is local minimum etc., for particular problem, its default algorithm may not restrain, and can adopt a kind of like this algorithm of optimization:
, in formula,
modified weight value when being n-th iteration,
for speedup factor,
for factor of momentum.According to the study,
,
time, convergence of algorithm speed, can avoid beating back and forth of when error surface is sleeve configuration algorithm.
Second step, specifies its input/output argument.When for parameter identification, neural network is used for approaching of nonlinear function.The number of the complex nature of the problem and input, output variable, and potential funtcional relationship between them is relevant.For a specific output, when choosing input, its input variable should be certain influence factor of output variable or have certain mapping relations.
The present invention set up Elman neural network be with rotor resistance Rr for export target, to the principal element that rotor resistance has an impact, rotor current, working time, frequency and environment temperature can be thought.Heating be one dynamically, the non-linear process of Great inertia, electric current is larger, frequency is higher, the working time, more long hair heat was larger, and the temperature rise that this heating causes determines the temperature of rotor together with environment temperature, and temperature is higher, and Rr is larger.Although environment temperature difference is comparatively large, motor is after work several hours, and usually all can reach thermal equilibrium state, temperature no longer rises.If using motor operating time as one of input variable, after motor radiating and environment reach thermal equilibrium, the increase of working time just no longer affects the change of rotor resistance.Four input parameters in resistance variations funtcional relationship, in fact become and only had 3 inputs, another factor just becomes disturbance variable, and in this, as identification input, the precision of result just can not be guaranteed.
Affect resistance variations to consider from temperature rise factor merely, total electric machine temperature rise can be reflected on motor housing, electric motor winding end temperature therefore can be selected to make a comprehensive input variable, its combined reaction total effect of each temperature influence factor.End temperature also can be used as the important evidence of For Diagnosing Faults of Electrical in addition, and while identification, also convenient have better monitoring disconnected to the running status of motor.
The change of rotor resistance mainly causes by the motor feels hot, its resistance variation with temperature roughly coincidence formula
, wherein, R20 is the resistance when environment temperature is 20 DEG C, and R (t) is resistance when temperature is t, and temperature-coefficient of electrical resistance when α 20 is 20 DEG C for its value of copper product is
, then its resistance varying-ratio of copper cash electrical resistance temperature variation as shown in Figure 3.
When holding warm instantaneous value constant, other input factor also can affect the size of rotor resistance value.Wherein be apparent that slip frequency the most.When motor stabilizing runs, the frequency of rotor current is slip frequency substantially, about
about Hz, the kelvin effect caused thus is less, generally can ignore.And in motor starting process, or during band heavy duty, slip frequency is comparatively large, so the impact of kelvin effect is quite serious, and can up to arriving
doubly.It is a lot of that this will cause rotor resistance to increase, and this phenomenon, with in the mouse cage asynchronous machine of deep slot type structure, shows more obvious, become a very important factor.Find the change of its rate of change of rotor resistance by slip frequency by further investigation, represent with quadratic function matching
, wherein s is slip frequency, then rotor resistance with slip frequency rate of change as shown in Figure 4.
3rd step, obtains training sample.Using above-mentioned input signal as master control condition, combine mutually with other input variable (change as slip frequency), adopt BP off-line algorithm to carry out training and obtain a collection of training sample, then rotor resistance model is set up using these data as reference, with the output of this model as the target of network to value, carry out error-duration model and modified weight.Embodiment is as follows:
A) set up BP network with the input and output parameter in step 3, as Fig. 5, input vector is
, hidden layer output vector:
, output layer output vector:
, desired output vector:
, the weight matrix between input layer to hidden layer represents with V,
, wherein column vector
for the weight vector that a hidden layer jth neuron is corresponding, the weight matrix between hidden layer to output layer represents with W,
, wherein column vector
for the weight vector that an output layer kth neuron is corresponding.
B) error E is determined.For output layer, have
,
, wherein
.For hidden layer, have
,
, wherein
.In above-mentioned function
be unipolarity Sigmoid function,
.When network exports and desired output does not wait, there is output error E, be defined as follows
, above error definition is expanded to hidden layer, has
, be expanded to input layer further, have
.
C) modified weight.From above-mentioned expression formula, network error originated from input is each layer weights
function, therefore adjust weights can change error E, obviously, adjustment weights principle be that error is constantly reduced, the adjustment amount of weights therefore should be made to be directly proportional to the Gradient Descent of error, namely
with
.
4th step, processes training sample.Because rotor resistance to be not only subject to the impact of this moment external environment condition and internal factor in the value in i moment, due to time stickiness, and to be subject to i-1, i-2 ..., the moment factors such as i-n impact, wish the value in M the moment with N number of data observation future in the past, desirable N number of adjacent sample is sliding window, and they are mapped as M value, be designed with certain overlapping sample data section.
Because the fluctuation range of each data is comparatively large, certain tittle does not even belong to an order of magnitude.Direct as input time, less input often can be flooded and not obvious by larger input, and while making the training time greatly increase, the precision that network exports also can reduce.Therefore when input too large or too little time, data inherent feature loss more.In order to avoid this situation, when inputting data of different nature, often input being normalized, making it entirely to drop on ± 1 scope.
When parameters normalization, relate to the select permeability of maximal value, quantification when this samples to AD is similar.By all inputs simultaneously divided by this quantized value, just all inputs can be limited to ± 1 scope within.Before this data prediction step should be placed on the input of neural network, if input is simulating signal, can consider together when AD changes.When data prediction, need to determine maximal value by theory calculate, observation estimation or experience.Embodiment is as follows:
When parameters normalization, relate to the select permeability of maximal value, quantification when this samples with AD is identical; By all inputs simultaneously divided by this quantized value, just all inputs can be limited to ± 1 scope within.
The quantification of temperature: can determine according to the class of insulation of motor, the temperature that usual motor runs larger than environment temperature to a certain extent time, namely reach thermal equilibrium state.Unless when fault occurs, just there will be some extreme temperature values.
The quantification of electric current: choosing of current maxima, can with reference to Design of Variable Frequency Speed Regulation index, and it is also that power device chooses reference.Need not be ultimate value unlike selection here, because the input of neural network is different from practical devices, there will not be the danger of burning device during overcurrent, abnormal mistakes high level be got rid of by inputting restriction.The maximum current of general threephase asynchronous machine, when appearing at starting, maximum starting current is about the 4-7 of its ratings doubly.
The quantification of slip frequency: when the maximal value of slip frequency normally appears at electric motor starting, considers that the time of this process is very little, and generally its maximum slip frequency in operational process desirable, can be taken as 25Hz.
5th step, carries out identification to rotor resistance.First using on its most significant temperature of impact and slip frequency as input, detect its effect with network.Add electric current again as input, whether the identification effect of supervising network improves, and finally determines the input parameter number of network.Embodiment is as follows:
For more significant influence factor, its inherent Nonlinear Mapping relation is comparatively strong, and Elman network hidden layer number can from less number.Hidden layer transport function selects tansig type, to make it to express the output in [-1,1] scope; Output layer function selects logsig type, and it is only positive enough through normalized resistance value for exporting.Rotor resistance is usual and little, and the motor that the present invention uses is substantially about 3 Europe, and more powerful motor, its rotor resistance value can be less than 1 Europe.In order to ensure the identification precision of neural network, the precision per mille during training of setting, namely error is 0.001.Higher precision can certainly be set, but limit due to the precision of real data sample phase, higher precision not too large meaning.
For the identification of rotor resistance, when arranging hidden neuron number and being 5, reach training objective and need 180 steps; When number is 6, need 138 steps; When number is 7, need 120 steps; When number is 8, need 137 steps; When number is 9, need 141 steps; When number is 10, need 171 steps.Increase with hidden neuron number, the train epochs that reaching target needs not reduces always, but has the trend of increase when being reduced to certain point.One is reasonably explained is after the structure of network has been enough to be input to the Nonlinear Mapping of output, along with neuron number object increases, need the weights of adjustment and feedback number to increase thereupon, needing the complexity of calculated amount and calculating to increase to reach same error size, therefore needing more step number to adjust.Express a better simply mapping with more neuron and just define certain redundancy, neuronicly do not reach maximum utilization.Although can increase the characteristic that network does interference in a sense, the complexity therefore increased is for inadvisable realizing its function with existing processor and electron device.Therefore for the present invention, hidden neuron number is selected to be 7.
Fig. 6 and Fig. 7 gives the output for test sample book of the network that trains and percentage error.Therefrom can find out, the emulation of network to input is better, its error of most of data point all remain on ± 1% within.In the end 5 input points, network output error starts significantly to increase, and particularly when the 50th point, its error reaches 4%, with regard to its error size, is enough to produce certain impact to high-precision control performance.
Increase electric current as after an input factor, the Nonlinear Mapping relation of constrained input complicates, therefore the scale of network must increase again.Increase gradually from the hidden neuron number 7 of above-mentioned two input networks, and consider according to the reduction speed speed and required train epochs of training time error, determine that hidden neuron number is 10, but train the step number generally needed all more than 200 steps.
Fig. 8 and Fig. 9 gives the network after increasing electric current input for the output of test sample book and percentage error.As can be seen from the figure, network export to the error of object vector most of data point all exceed ± 1%, ± 2%, ± 3% also very common, the data segment started, its error is more than ± 4%.Can find out that from percentage error figure identification exports and define many spikes, and saltus step back and forth, fluctuation range is larger.This point is similar to the feature of input current data, although have passed through filtering during current data sampling, but still level and smooth not, in protocone peak dot place waveform very irregular, be cause the main cause exporting saltus step back and forth.
What it should be noted that the representative of the data point comparatively is motor normality rotor resistance identification when slip frequency is less when running, and the impact at this moment caused by kelvin effect etc. is very little, and estimation occurs that such error can have an impact to control performance in long-time.Also can see, electric current reduces in the error of data point back segment to some extent as after input, but is transient process due to this section of representative, from the angle not too large impact that whole system performance improves.
Add electric current as input factor after, although the identification precision in these data points increases, consider that its precision have dropped on the whole, and this input makes the complexity of neural network structure increase on the contrary.In order to avoid simple problem is complicated, electric current is not inputted factor as one by the present invention.
Complete process from simple to complex like this, first to have the single parameter of input variable identification compared with definite relation, after its identification effect is compared and is verified, add other underlying factor gradually again, compare again, analyze, according to the actual effect of identification, finally obtain input parameter number and the so preferably network structure data of hidden neuron number.
Claims (3)
1., based on an induction electromotor rotor resistance parameter discrimination method for Elman neural network, it is characterized in that:
Step one, determine Elman neural network: the expression formula of Elman Neural Network Based Nonlinear state space is
;
;
, in formula: i, j, k represent input layer, hidden layer, the neuronic number of output layer respectively, and P, T represent input and output vector,
represent i-th weights be input between a jth hidden neuron,
represent the weights between a jth hidden neuron to kth output layer neuron,
represent the weights between a jth undertaking layer to a jth hidden neuron,
,
represent the transport function of hidden layer and output layer respectively,
,
the input represented respectively between each layer neuron is biased,
,
represent the feedback states vector of hidden layer output vector and its undertaking layer respectively, adopt a kind of like this algorithm of optimization:
, Elman neural network is optimized, in formula,
weighed value adjusting amount when being n-th iteration,
for speedup factor,
for factor of momentum;
Step 2, determine network structure: first, specify the input/output argument of Elman neural network, the principal element had an impact to rotor resistance is rotor current, frequency and environment temperature, make a comprehensive input variable to select electric motor winding end temperature; With rotor resistance
as output variable;
The acquisition of step 3, training sample: using the input variable in step 2 as master control condition, with turn electric current input variable, frequency input variable combines mutually, adopt BP off-line algorithm to carry out training and obtain a collection of training sample, rotor resistance model is set up as reference using these training sample data, with the output of this model as the target of network to value, carry out error-duration model and modified weight;
The process of step 4, training sample: process the training sample that step 3 obtains, when inputting data of different nature, is normalized input data, make it entirely to drop on ± 1 scope in, to facilitate training and the emulation of network;
Step 5, rotor resistance identification: adopt method from simple to complex, first to have the single parameter of input variable identification compared with definite relation, after its identification effect is compared and is verified, add other underlying factor gradually again, compare again, analyze, with the training sample handled well in step 4 for condition, identification is carried out to rotor resistance, first using on its most significant electric motor winding end temperature of impact and slip frequency as input, its effect is detected with network, add electric current again as input, whether the identification effect of supervising network improves, finally determine input parameter number and the hidden neuron number of network.
2. a kind of induction electromotor rotor resistance parameter discrimination method based on Elman neural network as claimed in claim 1, is characterized in that: described training sample acquisition methods is,
Step one, set up BP network with the input and output parameter in step 3: input vector is
, hidden layer output vector:
, output layer output vector:
, desired output vector:
, the weight matrix between input layer to hidden layer represents with V,
, wherein column vector
for hidden layer
qthe weight vector that individual neuron is corresponding, the weight matrix between hidden layer to output layer represents with W,
, wherein column vector
for output layer
sthe weight vector that individual neuron is corresponding;
Step 2, determine error E: for output layer, have
,
, wherein
, for hidden layer, have
,
, wherein
, in above-mentioned function
be unipolarity Sigmoid function,
, when network exports and desired output does not wait, there is output error E, be defined as follows
, above error definition is expanded to hidden layer, has
, be expanded to input layer further, have
;
Step 3, modified weight: from above-mentioned expression formula, network error originated from input is each layer weights
function, adjustment weights change error E, adjustment weights error is constantly reduced, the adjustment amount of weights is directly proportional, namely to the Gradient Descent of error
with
.
3. a kind of induction electromotor rotor resistance parameter discrimination method based on Elman neural network as claimed in claim 1, it is characterized in that: the disposal route of described training sample is, when parameters normalization, relate to the select permeability of maximal value, quantification when this samples with AD is identical; By all inputs simultaneously divided by this quantized value, all inputs be limited to ± 1 scope within, about the quantification of temperature input, the class of insulation according to motor is determined; Choosing about current maxima, adopts the maximum current of threephase asynchronous machine, and when appearing at starting, maximum starting current is 4-7 times of its ratings; The maximal value of slip frequency is when appearing at electric motor starting, the maximum slip frequency of power taking machine in operational process.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210463391.1A CN102937670B (en) | 2012-11-17 | 2012-11-17 | Method for identifying resistance parameters of rotors of induction motor on basis of Elman neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210463391.1A CN102937670B (en) | 2012-11-17 | 2012-11-17 | Method for identifying resistance parameters of rotors of induction motor on basis of Elman neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN102937670A CN102937670A (en) | 2013-02-20 |
| CN102937670B true CN102937670B (en) | 2015-05-27 |
Family
ID=47696580
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201210463391.1A Expired - Fee Related CN102937670B (en) | 2012-11-17 | 2012-11-17 | Method for identifying resistance parameters of rotors of induction motor on basis of Elman neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN102937670B (en) |
Families Citing this family (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103267826A (en) * | 2013-04-17 | 2013-08-28 | 沈阳大学 | Soft measurement method for online detection of gelatin concentration |
| CN103345546B (en) * | 2013-06-14 | 2016-09-07 | 国家电网公司 | The governor parameter discrimination method that frequency locus combines with particle cluster algorithm |
| CN107861061B (en) * | 2017-10-27 | 2019-11-01 | 安徽大学 | A kind of induction motor parameter on-line identification method of data-driven |
| CN109188185A (en) * | 2018-08-17 | 2019-01-11 | 中国大唐集团科学技术研究院有限公司 | A kind of generator rotor interturn short-circuit early stage online test method |
| CN112819158B (en) * | 2021-02-05 | 2024-02-27 | 凌坤(南通)智能科技有限公司 | Gas identification method based on optimized BP neural network |
| CN113556072B (en) * | 2021-07-29 | 2022-11-04 | 西南大学 | Online identification method for ultra-high speed motor parameters |
| CN115085627A (en) * | 2022-08-22 | 2022-09-20 | 成都微精电机股份公司 | Motor parameter dynamic identification method |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5023833A (en) * | 1987-12-08 | 1991-06-11 | California Institute Of Technology | Feed forward neural network for unary associative memory |
| CN101551663A (en) * | 2009-05-08 | 2009-10-07 | 华东理工大学 | Cracking reaction pre-warning method for cracking furnace |
-
2012
- 2012-11-17 CN CN201210463391.1A patent/CN102937670B/en not_active Expired - Fee Related
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5023833A (en) * | 1987-12-08 | 1991-06-11 | California Institute Of Technology | Feed forward neural network for unary associative memory |
| CN101551663A (en) * | 2009-05-08 | 2009-10-07 | 华东理工大学 | Cracking reaction pre-warning method for cracking furnace |
Non-Patent Citations (3)
| Title |
|---|
| 基于人工神经网络的感应电动机转子电阻参数辨识;王丽敏等;《农业工程学报》;19991231;第15卷(第4期);全文 * |
| 基于神经网络的无速度传感器感应电机转子电阻辨识方案;王涛;《交流技术与电力牵引》;20071231;全文 * |
| 基于神经网络的转子电阻在线辨识及补偿方法;徐小增等;《电工技术学报》;20021031;第17卷(第5期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN102937670A (en) | 2013-02-20 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN102937670B (en) | Method for identifying resistance parameters of rotors of induction motor on basis of Elman neural network | |
| CN102375120B (en) | Identify equipment and the method for AC asynchronous motor mechanical property parameter | |
| Nyanteh et al. | Application of artificial intelligence to stator winding fault diagnosis in Permanent Magnet Synchronous Machines | |
| US4442393A (en) | Apparatus and method for determining the operation of an induction motor | |
| Aliprantis et al. | Genetic algorithm-based parameter identification of a hysteretic brushless exciter model | |
| CN107451364B (en) | Method for identifying equivalent parameters of DFIG wind power plant | |
| CN105305411A (en) | System and method for a load anticipation feature and its tuning method for a generating set | |
| CN103684183B (en) | Rotational Speed of Asynchronous Motor method of estimation | |
| CN103760768A (en) | Generator set multivariable system identification method based on normal operating data | |
| CN104635028A (en) | Systems and methods for monitoring leakage current of industrial machine | |
| CN108282126B (en) | Speedless sensor current-sharing control method for double three-phase linear induction machines | |
| Rengifo et al. | Parameter estimation method for induction machines using the instantaneous impedance during a dynamic start-up | |
| CN106681133B (en) | A kind of Hydropower Unit model refinement type subspace closed-loop identification method | |
| Baccarini et al. | Sliding mode observer for on-line broken rotor bar detection | |
| Horen et al. | Simple mechanical parameters identification of induction machine using voltage sensor only | |
| Feltes et al. | Deriving model parameters from field test measurements | |
| Radhi et al. | Faults diagnosis in stator windings of high speed solid rotor induction motors using fuzzy neural network | |
| CN110838725B (en) | Parameter setting method and device for wind power plant closed-loop PI controller | |
| CN109039208A (en) | A kind of switched reluctance machines incremental inductance characteristic online test method | |
| CN105337547A (en) | Rotation speed identification method of asynchronous motor speed sensor-free vector control system | |
| WO2013000654A2 (en) | Control device and method for controlling an electric machine | |
| CN106849773A (en) | A kind of permagnetic synchronous motor servo inertia system identifying method | |
| CN114070146A (en) | Fault detection method, device, equipment and storage medium | |
| CN106712621B (en) | A kind of device and method recognizing induction motor stator resistance | |
| CN113193589B (en) | DFIG wind power plant subsynchronous oscillation suppression method based on digital twin simulation |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20150527 Termination date: 20151117 |
|
| EXPY | Termination of patent right or utility model |