CN114237043A - Gas turbine equipment transfer function closed-loop identification method based on deep learning - Google Patents
Gas turbine equipment transfer function closed-loop identification method based on deep learning Download PDFInfo
- Publication number
- CN114237043A CN114237043A CN202111429898.0A CN202111429898A CN114237043A CN 114237043 A CN114237043 A CN 114237043A CN 202111429898 A CN202111429898 A CN 202111429898A CN 114237043 A CN114237043 A CN 114237043A
- Authority
- CN
- China
- Prior art keywords
- model
- transfer function
- parameter
- loop
- gas turbine
- 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
- 230000006870 function Effects 0.000 title claims abstract description 39
- 238000012546 transfer Methods 0.000 title claims abstract description 33
- 238000000034 method Methods 0.000 title claims abstract description 27
- 238000013135 deep learning Methods 0.000 title claims abstract description 12
- 230000004044 response Effects 0.000 claims abstract description 30
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 13
- 238000012549 training Methods 0.000 claims abstract description 6
- 238000004088 simulation Methods 0.000 claims abstract description 4
- 230000003094 perturbing effect Effects 0.000 claims abstract description 3
- 238000002474 experimental method Methods 0.000 claims description 9
- 230000004913 activation Effects 0.000 claims description 6
- 238000011176 pooling Methods 0.000 claims description 4
- 230000003321 amplification Effects 0.000 claims description 3
- 238000003062 neural network model Methods 0.000 claims description 3
- 238000003199 nucleic acid amplification method Methods 0.000 claims description 3
- 238000010606 normalization Methods 0.000 claims description 2
- 230000008859 change Effects 0.000 abstract description 2
- 239000007789 gas Substances 0.000 description 16
- 238000012545 processing Methods 0.000 description 4
- 238000012360 testing method Methods 0.000 description 3
- 238000013528 artificial neural network Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- VNWKTOKETHGBQD-UHFFFAOYSA-N methane Chemical compound C VNWKTOKETHGBQD-UHFFFAOYSA-N 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 1
- 239000000446 fuel Substances 0.000 description 1
- 239000003345 natural gas Substances 0.000 description 1
- 238000010248 power generation Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B13/00—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion
- G05B13/02—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric
- G05B13/04—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators
- G05B13/042—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators in which a parameter or coefficient is automatically adjusted to optimise the performance
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Automation & Control Theory (AREA)
- Feedback Control In General (AREA)
Abstract
The invention discloses a deep learning-based closed-loop identification method for a transfer function of gas turbine equipment, which comprises the following steps of: (1) establishing a gas turbine key device as a first-order inertia pure lag FOPDT model, and establishing an open-loop transfer function model by adopting a PI controller as a feedback controller; (2) regularizing the model parameters, and perturbing the obtained dimensionless parameters; (3) closed loop step response simulation is carried out by utilizing a perturbation model generated by random numbers; (4) and constructing a deep learning convolutional neural network by taking the nominal time parameter and the closed loop step response curve as input parameters and taking the dimensionless model parameter as an output parameter, and training the convolutional neural network by using the numerical sequence. The closed-loop identification method is not limited by field operation conditions, and an open-loop transfer function is obtained under the closed-loop condition; the number of training samples is large enough to fully cover the uncertain change range of the actual object, and the identification accuracy is high.
Description
Technical Field
The invention relates to the field of thermal automatic control, in particular to a gas turbine equipment transfer function closed-loop identification method based on deep learning.
Background
At present, the power generation of a gas turbine taking natural gas as fuel becomes an important technology of power supply, and with the wide use of a gas-steam combined cycle generator set, the research on the operating characteristics and the control of the gas turbine is more and more. The gas turbine unit cannot perform an open-loop disturbance experiment due to the limitation of field operation conditions, and the development of closed-loop identification becomes an important technical means for designing a control system for obtaining the dynamic operation characteristics of the gas turbine.
System identification is an important part of modern control theory, and is to determine a system equivalent to the identified system based on input and output observations. Classical system identification methods include a step response method, an impulse response method, a least square method and derivatives thereof, but these methods are difficult to handle time lag and cannot determine the process identification of the time lag. Therefore, in order to solve the disadvantages of the classical identification method, applying new technologies to solve the problem becomes the focus of the current research.
Disclosure of Invention
The purpose of the invention is as follows: in view of the above problems, the present invention aims to provide a method for identifying a closed loop of a transfer function of a gas turbine equipment based on deep learning.
The technical scheme is as follows: the invention discloses a deep learning-based closed-loop identification method for a transfer function of gas turbine equipment, which comprises the following steps of:
(1) establishing a gas turbine key device as a first-order inertia pure lag FOPDT model, and establishing an open-loop transfer function model by adopting a PI controller as a feedback controller;
(2) regularizing the model parameters, and perturbing the obtained dimensionless parameters;
(3) closed loop step response simulation is carried out by utilizing a perturbation model generated by random numbers, and numerical value sequences of closed loop step response corresponding to each group of dimensionless parameters are recorded;
(4) and constructing a deep learning convolutional neural network by taking the nominal time parameter and the closed loop step response curve as input parameters and taking the dimensionless model parameter as an output parameter, and training the convolutional neural network by using the numerical sequence to obtain a neural network model of the identification model.
Further, the first-order inertia pure lag FOPDT model transfer function expression in step (1) is as follows:
where s is the complex frequency, K is the real object amplification factor, T is the real object time constant, L is the real object delay time, and parameter K, T, L is the parameter to be identified;
taking the PI controller as a feedback controller, wherein the corresponding transfer function expression is as follows:
wherein, KpIs the proportionality coefficient, TiIs the integral coefficient, Kp、TiThe value is adjusted according to the object characteristics and the control target;
thus, the open loop transfer function of the gas turbine control system is represented as:
according to the internal model setting method, the parameter K in the PI controllerp、TiSetting, wherein the setting expression is as follows:
Ti＝T0
wherein K0For the magnification factor, T, of the object model0Is the object model time constant, L0Is the object model delay time.
Further, the parameter regularization in step (2) is defined as follows:
beta, gamma, alpha are dimensionless parameters with nominal values of 1, theta0Is a nominal time parameter, theta is a time parameter to be solved;
by substituting the above equation into the gas turbine control system, the open loop transfer function is rewritten as:
further, the perturbation in the step (2) is indicated at a known nominal time parameter theta0Under the condition, random perturbation of + -30% is carried out on dimensionless numbers beta, gamma and alpha, and each dimensionless parameter is [0.7, 1.3 ]]The values are uniformly and randomly selected in the range.
Further, step (3) of carrying out a step response experiment on the closed-loop system on the basis of each group of random number pairs, carrying out the closed-loop step response experiment according to a corresponding group of beta, gamma and alpha, taking points on a curve with equal time step length delta g after the response curve tends to be stable, and recording each point and a nominal time parameter theta0And dimensionless numbers β, γ, α.
Further, the convolutional neural network comprises 1 input layer, 4 convolutional layers, 4 batch normalization layers, 4 activation function layers, 2 average pooling layers, one discarding layer, 1 fully-connected layer and 1 regression layer.
Has the advantages that: compared with the prior art, the invention has the following remarkable advantages: the closed-loop identification method is not limited by field operation conditions, and an open-loop transfer function can be obtained under the closed-loop condition; the number of training samples is large enough, the uncertain change range of the actual object can be fully covered, and the recognition accuracy is high; the problem that a first-order time lag model FOPDT is difficult to identify is solved; the parameter identification process is simple and easy to operate.
Drawings
FIG. 1 is a flow chart of the present invention;
FIG. 2 is a schematic diagram of an open-loop transfer function model structure;
FIG. 3 is a diagram of a convolutional neural network model;
FIG. 4 is a graph of absolute percentage error for each parameter, where graph (a) is a graph of absolute percentage error for parameter β; graph (b) is a plot of the absolute percent error of the parameter γ; graph (c) is a plot of the absolute percent error for parameter α;
fig. 5 is a closed loop step response graph.
Detailed Description
In this embodiment, a flowchart of a method for identifying a closed loop of a transfer function of a gas turbine plant based on deep learning is shown in fig. 1, and includes the following steps:
(1) establishing a gas turbine key device as a first-order inertia pure lag FOPDT model, wherein the key device is a valve, a pipeline, a rotating speed, a sensor and the like, an open-loop transfer function model is established by adopting a PI controller as a feedback controller, and the model structure is shown in figure 2;
the first-order inertia pure lag FOPDT model transfer function expression is as follows:
where s is the complex frequency, K is the real object amplification factor, T is the real object time constant, L is the real object delay time, and parameter K, T, L is the parameter to be identified;
taking the PI controller as a feedback controller, wherein the corresponding transfer function expression is as follows:
wherein, KpIs the proportionality coefficient, TiIs the integral coefficient, Kp、TiThe value is adjusted according to the object characteristics and the control target;
thus, the open loop transfer function of the gas turbine control system is represented as:
according to internal model settingParameter K in Fair PI controllerp、TiSetting, wherein the setting expression is as follows:
Ti＝T0
wherein K0For the magnification factor, T, of the object model0Is the object model time constant, L0For object model delay time, K0，T0，L0Obtained from model prior knowledge.
(2) Regularizing the model parameters, and defining the formula as follows: (ii) a
Beta, gamma, alpha are dimensionless parameters with nominal values of 1, theta0Is a nominal time parameter, theta is a time parameter to be solved;
by substituting the above equation into the gas turbine control system, the open loop transfer function is rewritten as:
at a known nominal time parameter theta0Under the condition, random perturbation of +/-30% is carried out on dimensionless parameters of beta, gamma and alpha, and the beta, gamma and alpha are in [0.7, 1.3 ]]Uniformly and randomly taking values in the range, and taking the nominal time parameter as theta0∈[0.5，10]. The formula corresponding to each group of dimensionless parameters in the above formula represents a similar object, not a specific object.
(3) And (3) carrying out closed-loop step response simulation by using a perturbation model generated by random numbers, and recording a numerical sequence of closed-loop step response corresponding to each group of dimensionless parameters.
Performing step response experiment on the closed-loop system on the basis of each group of random number pairs, performing closed-loop step response experiment according to a corresponding group of beta, gamma and alpha, and waiting forAfter the response curve tends to be stable, points are taken on the curve at equal time step length delta g, and each point and a nominal time parameter theta are recorded0And dimensionless numbers β, γ, α.
(4) And constructing a deep learning convolutional neural network by taking the nominal time parameter and the closed loop step response curve as input parameters and taking the dimensionless model parameter as an output parameter, and training the convolutional neural network by using the numerical sequence to obtain a neural network model of the identification model.
The convolutional neural network has a structure comprising an input layer, a first convolution layer, a first batch processing layer, a first activation function layer, a first average pooling layer, a second convolution layer, a second batch processing layer, a second activation function layer, a second average pooling layer, a third convolution layer, a third batch processing layer, a third activation function layer, a fourth convolution layer, a fourth batch processing layer, a fourth activation function layer, a discarding layer, a full-connection layer and a regression layer which are sequentially arranged, and the specific structure is shown in figure 3.
The identification method is evaluated through parameter identification performance indexes, absolute percentage errors are used as the performance indexes, and the expression is as follows:
wherein k is ∈ [1, n ]]And k belongs to Z, n is the number of groups in the test set, mkRepresenting absolute percentage error, t, of the parameter of the kth groupkTrue value of parameter, p, of the k-th groupkIs the parameter identification value of the kth group.
The test set data has 1000 groups, that is, n is 1000, the absolute percentage error distribution of each parameter is counted, the distribution results obtained by β, γ, α are respectively shown in fig. 4(a), (b), (c), and more than 850 groups of the 1000 test sets have absolute percentage errors of the parameters within the range of (0, 3%).
As shown in fig. 5, at a nominal time parameter θ0Under the premise of 3.7435, when the true value of the dimensionless parameter is:
then, a closed loop step response experiment is carried out on the system to obtain a corresponding step response curve, and the response curve and a nominal time parameter theta are measured0As input parameters of the neural network, obtaining identification dimensionless parameters through the convolutional neural network:
and due to the parameter K0，T0，L0Obtaining K, T and L by model prior knowledge, namely obtaining parameters to be identified in the open-loop transfer function of the important equipment of the gas turbine according to the closed-loop step response curve and the nominal time parameter of the system, and finally obtaining the open-loop transfer function. And then, carrying out a closed loop step response experiment on the open loop transfer function corresponding to the identification parameter to obtain the identification step response curve shown in the figure 5, wherein the identification accuracy of the neural network established by the method is high according to the coincidence degree of the two response curves shown in the figure 5.
Claims (6)
1. The method for identifying the closed loop of the transfer function of the gas turbine equipment based on deep learning is characterized by comprising the following steps of:
(1) establishing a gas turbine key device as a first-order inertia pure lag FOPDT model, and establishing an open-loop transfer function model by adopting a PI controller as a feedback controller;
(2) regularizing the model parameters, and perturbing the obtained dimensionless parameters;
(3) closed loop step response simulation is carried out by utilizing a perturbation model generated by random numbers, and numerical value sequences of closed loop step response corresponding to each group of dimensionless parameters are recorded;
(4) and constructing a deep learning convolutional neural network by taking the nominal time parameter and the closed loop step response curve as input parameters and taking the dimensionless model parameter as an output parameter, and training the convolutional neural network by using the numerical sequence to obtain a neural network model of the identification model.
2. The method of claim 1, wherein the first order inertia pure lag FOPDT model transfer function expression in step (1) is:
where s is the complex frequency, K is the real object amplification factor, T is the real object time constant, L is the real object delay time, and parameter K, T, L is the parameter to be identified;
taking the PI controller as a feedback controller, wherein the corresponding transfer function expression is as follows:
wherein, KpIs the proportionality coefficient, TiIs the integral coefficient, Kp、TiThe value is adjusted according to the object characteristics and the control target;
thus, the open loop transfer function of the gas turbine control system is represented as:
according to the internal model setting method, the parameter K in the PI controllerp、TiSetting, wherein the setting expression is as follows:
Ti＝T0
wherein K0For the magnification factor, T, of the object model0Is the object model time constant, L0Is the object model delay time.
3. The gas turbine plant transfer function closed-loop identification method of claim 2, wherein the parameter regularization in step (2) is defined as follows:
s'＝L0s,
beta, gamma, alpha are dimensionless parameters with nominal values of 1, theta0Is a nominal time parameter, theta is a time parameter to be solved;
by substituting the above equation into the gas turbine control system, the open loop transfer function is rewritten as:
4. the method of claim 3, wherein the perturbation is determined at a known nominal time parameter θ for the perturbation in step (2)0Under the condition, random perturbation of + -30% is carried out on dimensionless numbers beta, gamma and alpha, and each dimensionless parameter is [0.7, 1.3 ]]The values are uniformly and randomly selected in the range.
5. The method for identifying the closed loop of the transfer function of the gas turbine equipment as claimed in claim 4, wherein the step (3) is to perform a step response experiment on the closed loop system based on each set of random number pairs, perform the closed loop step response experiment according to a corresponding set of β, γ, α, take points with an equal time step Δ g on the curve after the response curve tends to be stable, and record each point and the nominal time parameter θ0And dimensionless numbers β, γ, α.
6. The gas turbine plant transfer function closed-loop identification method of claim 1, wherein the convolutional neural network comprises one input layer, four convolutional layers, four batch normalization layers, four activation function layers, two average pooling layers, one discarding layer, one fully-connected layer, one regression layer.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111429898.0A CN114237043B (en) | 2021-11-29 | 2021-11-29 | Gas turbine equipment transfer function closed-loop identification method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111429898.0A CN114237043B (en) | 2021-11-29 | 2021-11-29 | Gas turbine equipment transfer function closed-loop identification method based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114237043A true CN114237043A (en) | 2022-03-25 |
| CN114237043B CN114237043B (en) | 2024-04-16 |
Family
ID=80751805
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111429898.0A Active CN114237043B (en) | 2021-11-29 | 2021-11-29 | Gas turbine equipment transfer function closed-loop identification method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114237043B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115291509A (en) * | 2022-06-24 | 2022-11-04 | 东南大学溧阳研究院 | Gas turbine dynamic parameter online identification method based on hybrid deep learning model |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPH04346102A (en) * | 1991-05-23 | 1992-12-02 | Hitachi Ltd | Pid parameter automatic tuning method |
| CN1487268A (en) * | 2003-08-12 | 2004-04-07 | 上海交通大学 | Multi-varible system structural colsed-loop distinguishing method based on step response test |
| CN108333933A (en) * | 2018-01-26 | 2018-07-27 | 东南大学 | A kind of single order pure delay system closed-loop identification method |
| CN110174841A (en) * | 2019-04-30 | 2019-08-27 | 杭州意能电力技术有限公司 | The parallel control method of industrial process large time delay inertia system based on parameter optimization |
| CN110222711A (en) * | 2019-04-30 | 2019-09-10 | 杭州意能电力技术有限公司 | A kind of multistage inertia system Open-loop Identification method of industrial process based on deep learning |
| CN110244557A (en) * | 2019-04-30 | 2019-09-17 | 国网浙江省电力有限公司电力科学研究院 | A kind of closed loop modeling method of the multistage inertia closed-loop system of industrial process |
-
2021
- 2021-11-29 CN CN202111429898.0A patent/CN114237043B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPH04346102A (en) * | 1991-05-23 | 1992-12-02 | Hitachi Ltd | Pid parameter automatic tuning method |
| CN1487268A (en) * | 2003-08-12 | 2004-04-07 | 上海交通大学 | Multi-varible system structural colsed-loop distinguishing method based on step response test |
| CN108333933A (en) * | 2018-01-26 | 2018-07-27 | 东南大学 | A kind of single order pure delay system closed-loop identification method |
| CN110174841A (en) * | 2019-04-30 | 2019-08-27 | 杭州意能电力技术有限公司 | The parallel control method of industrial process large time delay inertia system based on parameter optimization |
| CN110222711A (en) * | 2019-04-30 | 2019-09-10 | 杭州意能电力技术有限公司 | A kind of multistage inertia system Open-loop Identification method of industrial process based on deep learning |
| CN110244557A (en) * | 2019-04-30 | 2019-09-17 | 国网浙江省电力有限公司电力科学研究院 | A kind of closed loop modeling method of the multistage inertia closed-loop system of industrial process |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115291509A (en) * | 2022-06-24 | 2022-11-04 | 东南大学溧阳研究院 | Gas turbine dynamic parameter online identification method based on hybrid deep learning model |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114237043B (en) | 2024-04-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109147878B (en) | Soft measurement method for free calcium of cement clinker | |
| CN110689171A (en) | Turbine health state prediction method based on E-LSTM | |
| CN107436971B (en) | Improved Latin hypercube sampling method suitable for non-positive stereotype correlation control | |
| CN110287540B (en) | Photovoltaic cell parameter identification method based on elite crowd differential evolution algorithm | |
| CN112149903B (en) | Primary frequency modulation analysis and optimization method of thermal power generating unit based on BP neural network algorithm | |
| CN109800449B (en) | Neural network-based aeroengine compression component characteristic correction method | |
| CN111006843B (en) | Continuous variable speed pressure method of temporary impulse type supersonic wind tunnel | |
| CN114548368A (en) | Modeling method, prediction model and prediction method of lithium battery temperature field prediction model based on multilayer nuclear overrun learning machine | |
| CN112836884A (en) | Copula-DBiLSTM-based method for accurately predicting multi-element load of comprehensive energy system | |
| CN114462718A (en) | CNN-GRU wind power prediction method based on time sliding window | |
| CN107272412A (en) | A kind of identifying approach of intermittent wind tunnel flow field control | |
| CN114237043A (en) | Gas turbine equipment transfer function closed-loop identification method based on deep learning | |
| CN113283004A (en) | Aero-engine degradation state fault diagnosis method based on transfer learning | |
| CN112669168A (en) | Short-term wind power prediction method | |
| CN109056056A (en) | A kind of Modelling of Crystal Growth in CZ-Si Pulling process thermal field identification Method | |
| CN114626573A (en) | Load prediction method for optimizing extreme learning machine based on improved multivariate universe algorithm | |
| CN118071205A (en) | Self-adaptive control system for raw material like coke for lithium battery cathode material | |
| CN110276478B (en) | Short-term wind power prediction method based on segmented ant colony algorithm optimization SVM | |
| CN117408833A (en) | Short-term load prediction method based on load identification | |
| CN116757078A (en) | Method and system for measuring flow velocity of pulverized coal based on acting force | |
| Ma et al. | Prediction of thermal system parameters based on PSO-ELM hybrid algorithm | |
| CN116292057A (en) | Control method, device, equipment and medium for variable-speed pumped storage unit | |
| CN105117530B (en) | Method for identifying parameters of steam turbine and speed regulating system thereof by combination of thickness and thickness regulation | |
| CN115291509A (en) | Gas turbine dynamic parameter online identification method based on hybrid deep learning model | |
| Li et al. | A wind power prediction model based on optimized N-BEATS network with multivariate inputs |
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 |