CN114326406A - Coordination control method based on vector machine online identification prediction - Google Patents
Coordination control method based on vector machine online identification prediction Download PDFInfo
- Publication number
- CN114326406A CN114326406A CN202111669703.XA CN202111669703A CN114326406A CN 114326406 A CN114326406 A CN 114326406A CN 202111669703 A CN202111669703 A CN 202111669703A CN 114326406 A CN114326406 A CN 114326406A
- Authority
- CN
- China
- Prior art keywords
- model
- prediction
- error
- control
- data set
- 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
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention belongs to the field of automatic control of thermal power generating units, and particularly relates to a coordination control method based on vector machine online identification prediction; the technical scheme is as follows: s1, acquiring and processing the online identification model data; s2, structural design and parameter optimization of the online identification model; s3, designing and calculating the output of the prediction controller; s4, repeating the steps from S1 to S3 at the next time point K + 1; the thermal power generating unit coordination control method solves the problem of mismatch between the transfer function model parameters and the actual object model, and fundamentally improves the control quality of a coordination control system in the process of participating in peak shaving of the thermal power generating unit.
Description
Technical Field
The invention belongs to the field of automatic control of thermal power generating units, and particularly relates to a coordination control method based on vector machine online identification prediction.
Background
With the increasingly prominent global energy crisis and environmental protection problems, the global power development is moving towards the direction of low-carbon development, and in recent years, new energy is developed in China on a large scale, and in order to solve the problem of power generation and grid connection of a new energy unit, the participation of a thermal power unit in power grid deep peak shaving has become a normalized requirement.
In the peak shaving process of the thermal power generating unit, the response speed of the thermal power generating unit is greatly limited due to the differentiated dynamic response characteristics of the two sides of the boiler and the steam turbine, and a new control strategy research needs to be carried out for improving the response speed of the thermal power generating unit to the middle-shaving AGC instruction and improving the economy of the thermal power generating unit.
In the unit variable load process, on one hand, the characteristics of a controlled object in the unit furnace coordination control system continuously change, and the existing transfer function model parameters identified based on the rated working condition are mismatched with an actual object model, so that the control quality of the coordination control system is deteriorated; on the other hand, the response deviation of the unit in the variable load process is aggravated by the conventional fixed parameter PID control strategy, and in conclusion, under the existing control strategy, a satisfactory control effect is difficult to obtain in the unit deep peak shaving process.
Disclosure of Invention
The invention overcomes the defects in the prior art, and provides the thermal power unit coordination control method which can ensure that the thermal power unit boiler coordination system has better control quality and robustness in the unit wide load peak shaving process by using the support vector machine.
In order to solve the technical problems, the invention adopts the technical scheme that: a coordination control method based on vector machine online identification prediction is realized by the following steps:
s1, acquiring and processing the online identification model data;
setting a time K aiming at a main steam pressure control loop in a coordinated control system, and acquiring the input of a control modelA data set u (k) and an output data set y (k), wherein the sampling data set is set to be L groups to obtain an identification data set TL(ii) a Identifying data set TLA rolling time window determination is employed, which is expressed as:
TL＝{(uk-L+1,yk-L+1),(uk-L+2,yk-L+2),...,(uk,yk)},u∈R,y∈R
normalizing input data and output data used for model identification to an interval of [ -1,1] by adopting the following expression:
in the formula, x and x' represent values before and after data normalization, respectively.
Normalizing the input data set u (k) and the output data set y (k) of the control model;
randomly dividing the normalized data set to determine a model training data set and a model verification data set;
s2, structural design and parameter optimization of the online identification model;
giving initial values of a support vector machine model parameter hyperparametric error penalty factor gamma and a Gaussian kernel width parameter sigma;
searching optimal values of a parameter penalty factor gamma and a Gaussian kernel width parameter sigma by using the training data set and adopting a genetic algorithm, and verifying the generalization capability of the model through the verification data set;
s3, designing and calculating the output of the prediction controller;
setting a prediction time domain P and a control time domain M, and calculating the output of a prediction model;
calculating the error e between the system output value and the model output valuem(k)；
Using said error em(k) Feedback correction is carried out on the output of the prediction model to obtain a corrected output value of the prediction model;
calculating a control sequence according to a quadratic optimization target of the output error and the control increment;
and S4, repeating the steps from S1 to S3 at the next time point K + 1.
The online identification model in the step S2 adopts a least square support vector machine model, and the calculation formula of the least square support vector machine model is as follows:
for a given datum D { (x)i,yi):i＝1,2,...,l}，
Wherein x isi∈RnIs an n-dimensional input vector, yiE is R as a target output vector;
by non-linearly mapping phi (x), yi＝WTφ(xi)+b，W∈Rn、b∈R (1-1)；
According to the principle of solving the target and minimizing the structural risk, the following conditions are satisfied in the above formula (1-1):
|yi-WTφ(xi)-b|≤X
defining an error loss function as a quadratic term of the error
The above formula (1-2) can be represented as:
optimizing the error problem:
in the formula, alphaiAnd e.g. R is a Lagrange multiplier, and the above formula (1-4) is optimized to obtain:
eliminating W, e, the following matrix equation can be obtained:
wherein:
y＝[y1,y2,...,yl]T
l＝[1,1,...,1]T∈Rl
α＝[α1,α2,...,αl]
Ω＝{Ωij}l×l,Ωij＝φT(xi)φ(xj)＝K(xi,xj)
where Ω is a symmetric matrix and K (,) is a kernel function.
The least squares support vector machine prediction function is represented as:
the radial basis function, as a kernel function in the prediction function, is expressed as follows:
where σ is a gaussian kernel width parameter.
In the process of searching the optimal values of the parameter penalty factor gamma and the Gaussian kernel width parameter sigma by the genetic algorithm, the fitness function is as follows:
the evaluation function of the generalization ability of the model adopts root mean square error and is expressed as follows:
the single-step prediction model calculation formula of the least square support vector machine at the time K in the step S3 is as follows:
ym(k+1)＝fLS-SVM[yp(k),yp(k-1),u(k)] (2-1)
performing multi-step iteration on the basis of single-step prediction to realize multi-step prediction;
ym(k+j)＝fLS-SVM[yp(k+j-1),yp(k+j-2),u(k+j-1)] (2-2)
j 1,2, P denotes a prediction time domain;
wherein u (k + M-1) ═ u (k + M) · u (k + P-1) (2-3)
M represents a control time domain, and M is less than or equal to P.
Calculating the error e between the system output value and the model output valuem(k)，
em(k)＝y(k)-ym(k) (2-4)；
Using error em(k) For the predicted output ym(k + j) feedback correction is performed, and the corrected prediction model output value is expressed as follows:
yp(k+j)＝ym(k+j)+he(k)，j＝1,2,...,P (2-5)
in the formula, h is an error correction coefficient;
finally, calculating a control sequence according to the quadratic optimization target of the output error and the control increment;
the quadratic objective function of the rolling optimization is expressed as follows:
in the formula, λiNon-negative weighting coefficient, y, for the control quantityr(k + i) is a reference trajectory;
the reference trajectory is generally defined as an exponential curve, expressed as follows:
in the formula (I), the compound is shown in the specification,
representing a smoothing factor, r representing an input to the system;
at the current k moment, a set of optimal control sequences is found from the allowable interval of the controlled variable by a quadratic optimization objective function
U, U ═ U (k), U (k + 1.., U (k + M-1) ], minimizes the objective function J.
The constraint of the equations (1-3) is yi＝WTφ(xi)+b+ei，i＝1,2,...,l。
Compared with the prior art, the invention has the beneficial effects that:
1. the thermal power generating unit coordinated control method solves the problem that the control quality of a coordinated control system is poor due to mismatching of the transfer function model parameters and the actual object model, and fundamentally improves the control quality of the coordinated control system in the process that the thermal power generating unit participates in peak shaving.
2. The method has the characteristics of small modeling sample, nonlinearity and simple calculation of the support vector machine, realizes online correction of the prediction model based on the genetic algorithm, and realizes the control optimization of the whole working condition of the unit.
Drawings
The invention is further described below with reference to the accompanying drawings.
Fig. 1 is a control schematic block diagram of the present invention.
Detailed Description
As shown in fig. 1, step one: data acquisition and processing for online identification models
And acquiring input and output data of the control model aiming at a main steam pressure control loop in the coordinated control system.
Let the current time point be k time point, and the input and output of the model are respectively represented by u (k) and y (k). And determining a model identification data set by adopting a rolling time window, and if a sampling data set is an L set, representing the identification data set as follows:
TL＝{(uk-L+1,yk-L+1),(uk-L+2,yk-L+2),...,(uk,yk)},u∈R,y∈R
normalizing the collected data set in the rolling time window, wherein the input and output data used by model identification are normalized to the range of [ -1,1] by adopting the following expression:
in the formula, x and x' represent values before and after data normalization, respectively.
And randomly dividing the normalized data set to determine a model training data set and a model verification data set. And randomly disordering the sequence pointed by the sample data set to generate a new sequence number, and sequentially selecting a training set and a verification set which meet the set number of samples from the new sequence number sample set.
Step two: and (5) carrying out online identification model structure design and parameter optimization.
In the method, an online identification model adopts a least square support vector machine model with strong generalization capability based on a small sample, and the least square support vector machine model is designed as follows:
for a given datum D { (x)i,yi) I 1, 2.. gtorel } wherein, xi∈RnIs an n-dimensional input vector, yiAnd e.R is the target output vector. And transforming the nonlinear estimation problem into a high-dimensional feature space linear function estimation problem by mapping the input vector from the input space to the feature space through nonlinear mapping phi (x), wherein the linear function is expressed as follows:
yi＝WTφ(xi)+b，W∈Rn，b∈R (1-1)
according to the principle of solving the target and minimizing the structured risk, the following conditions are satisfied by the above formula:
|yi-WTφ(xi)-b|≤X
defining an error loss function as a quadratic term of the error
The above formula can be represented as
the equality constraints of the above equation are: y isi＝WTφ(xi)+b+ei，i＝1,2,...,l
Solving the optimization problem by using a Lagrange method, wherein the Lagrange function is quoted as follows:
in the formula, alphaiE, taking R as Lagrange multiplier, and optimizing the formula to obtain:
eliminating W, e, the following matrix equation can be obtained:
wherein:
y＝[y1,y2,...,yl]T
l＝[1,1,...,1]T∈Rl
α＝[α1,α2,...,αl]
Ω＝{Ωij}l×l,Ωij＝φT(xi)φ(xj)＝K(xi,xj)
where Ω is a symmetric matrix and K (,) is a kernel function.
The least squares support vector machine prediction function is represented as:
the radial basis function is used herein as a kernel function in the prediction function, and is expressed as follows:
where σ is a gaussian kernel width parameter.
In a support vector machine model, an error penalty factor gamma and a Gaussian kernel width parameter sigma greatly influence the performance of the model, a training data set is utilized, a genetic algorithm is adopted to search the optimal values of the two parameters, and the generalization capability of the model is verified through a verification data set.
In the genetic algorithm parameter optimizing process, the fitness function adopts the following form:
the evaluation function of the generalization ability of the model adopts root mean square error and is expressed as follows:
step three: predictive controller design and output calculation
The output of the predictive model is first calculated. In the control method, a least square support vector machine model is used as a prediction model, and a single-step prediction model mathematical expression based on the least square support vector machine at the time k is as follows:
ym(k+1)＝fLS-SVM[yp(k),yp(k-1),u(k)] (2-1)
and performing multi-step iteration on the basis of single-step prediction to realize multi-step prediction.
ym(k+j)＝fLS-SVM[yp(k+j-1),yp(k+j-2),u(k+j-1)] (2-2)
j 1,2, P denotes a prediction time domain;
wherein u (k + M-1) ═ u (k + M) · u (k + P-1) (2-3)
M represents a control time domain, and M is less than or equal to P.
Then calculating the error e between the system output value and the model output valuem(k)，em(k)＝y(k)-ym(k)。
Using the error pair to predict output ym(k + j) feedback correction is performed, and the corrected prediction model output value is expressed as follows:
yp(k+j)＝ym(k+j)+he(k)，j＝1,2,...,P
in the formula, h is an error correction coefficient;
and finally, calculating a control sequence according to the quadratic optimization target of the output error and the control increment.
The quadratic objective function of the rolling optimization is expressed as follows:
in the formula, λiNon-negative weighting coefficient, y, for the control quantityr(k + i) is a reference trajectory.
The reference trajectory is generally defined as an exponential curve, expressed as follows:
in the formula (I), the compound is shown in the specification,
representing the smoothing factor and r representing the input to the system.
At the current time k, a set of optimal control sequences U is found from the control quantity tolerance interval by using a quadratic optimization objective function, wherein the optimal control sequences U are [ U (k), U (k + 1..), U (k + M-1) ] enable the objective function J to be minimum.
Step four: and returning to the step one when the next time point k +1 is the time point.
The above embodiments are merely illustrative of the principles of the present invention and its effects, and do not limit the present invention. It will be apparent to those skilled in the art that modifications and improvements can be made to the above-described embodiments without departing from the spirit and scope of the invention. Accordingly, it is intended that all equivalent modifications or changes be made by those skilled in the art without departing from the spirit and technical spirit of the present invention, and be covered by the claims of the present invention.
Claims (6)
1. A coordination control method based on vector machine online identification prediction is characterized by comprising the following steps:
s1, acquiring and processing the online identification model data;
setting a time K aiming at a main steam pressure control loop in a coordinated control system, acquiring an input data set u (K) and an output data set y (K) of the control model, setting a sampling data set into an L group, and obtaining an identification data set TL；
Normalizing the input data set u (k) and the output data set y (k) of the control model;
randomly dividing the normalized data set to determine a model training data set and a model verification data set;
s2, structural design and parameter optimization of the online identification model;
giving initial values of a support vector machine model parameter hyperparametric error penalty factor gamma and a Gaussian kernel width parameter sigma;
searching optimal values of a parameter penalty factor gamma and a Gaussian kernel width parameter sigma by using the training data set and adopting a genetic algorithm, and verifying the generalization capability of the model through the verification data set;
s3, designing and calculating the output of the prediction controller;
setting a prediction time domain P and a control time domain M, and calculating the output of a prediction model;
calculating the error e between the system output value and the model output valuem(k)；
Using said error em(k) Feedback correction is carried out on the output of the prediction model to obtain a corrected output value of the prediction model;
calculating a control sequence according to a quadratic optimization target of the output error and the control increment;
and S4, repeating the steps from S1 to S3 at the next time point K + 1.
2. The method as claimed in claim 1, wherein the online identification model in step S2 adopts a least square support vector machine model, and the calculation formula of the least square support vector machine model is as follows:
for a given datum D { (x)i,yi):i＝1,2,...,l}，
Wherein x isi∈RnIs an n-dimensional input vector, yiE is R as a target output vector;
by non-linearly mapping phi (x), yi＝WTφ(xi)+b，W∈Rn、b∈R (1-1)；
According to the principle of solving the target and minimizing the structural risk, the following conditions are satisfied in the above formula (1-1):
|yi-WTφ(xi)-b|≤X
defining an error loss function as a quadratic term of the error
The above formula (1-2) can be represented as:
optimizing the error problem:
in the formula, alphaiAnd e.g. R is a Lagrange multiplier, and the above formula (1-4) is optimized to obtain:
eliminating W, e, the following matrix equation can be obtained:
wherein:
y＝[y1,y2,...,yl]T
l＝[1,1,...,1]T∈Rl
α＝[α1,α2,...,αl]
Ω＝{Ωij}l×l,Ωij＝φT(xi)φ(xj)＝K(xi,xj)
where Ω is a symmetric matrix and K (,) is a kernel function.
The least squares support vector machine prediction function is represented as:
the radial basis function, as a kernel function in the prediction function, is expressed as follows:
where σ is a gaussian kernel width parameter.
3. The coordination control method based on vector machine online identification prediction as claimed in claim 2, wherein in the process of searching the optimal values of the parameter penalty factor γ and the gaussian kernel width parameter σ by the genetic algorithm, the fitness function is as follows:
the evaluation function of the generalization ability of the model adopts root mean square error and is expressed as follows:
4. the method as claimed in claim 1, wherein the calculation formula of the single-step prediction model of the least-squares support vector machine at time K in step S3 is as follows:
ym(k+1)＝fLS-SVM[yp(k),yp(k-1),u(k)] (2-1)
performing multi-step iteration on the basis of single-step prediction to realize multi-step prediction;
ym(k+j)＝fLS-SVM[yp(k+j-1),yp(k+j-2),u(k+j-1)] (2-2)
j 1,2, P denotes a prediction time domain;
wherein u (k + M-1) ═ u (k + M) · u (k + P-1) (2-3)
M represents a control time domain, and M is less than or equal to P.
5. The method of claim 4, wherein the error e between the system output value and the model output value is calculatedm(k)，
em(k)＝y(k)-ym(k) (2-4)；
Using error em(k) For the predicted output ym(k + j) feedback correction is performed, and the corrected prediction model output value is expressed as follows:
yp(k+j)＝ym(k+j)+he(k)，j＝1,2,...,P (2-5)
in the formula, h is an error correction coefficient;
finally, calculating a control sequence according to the quadratic optimization target of the output error and the control increment;
the quadratic objective function of the rolling optimization is expressed as follows:
in the formula, λiNon-negative weighting coefficient, y, for the control quantityr(k + i) is a reference trajectory;
the reference trajectory is generally defined as an exponential curve, expressed as follows:
in the formula (I), the compound is shown in the specification,
representing a smoothing factor, r representing an input to the system;
at the current time k, a set of optimal control sequences U is found from the control quantity tolerance interval by using a quadratic optimization objective function, wherein the optimal control sequences U are [ U (k), U (k + 1..), U (k + M-1) ] enable the objective function J to be minimum.
6. The method for coordination control based on vector machine online identification prediction as claimed in claim 2, wherein the constraint condition of formula (1-3) is yi＝WTφ(xi)+b+ei，i＝1,2,...,l。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111669703.XA CN114326406A (en) | 2021-12-31 | 2021-12-31 | Coordination control method based on vector machine online identification prediction |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111669703.XA CN114326406A (en) | 2021-12-31 | 2021-12-31 | Coordination control method based on vector machine online identification prediction |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114326406A true CN114326406A (en) | 2022-04-12 |
Family
ID=81021579
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111669703.XA Pending CN114326406A (en) | 2021-12-31 | 2021-12-31 | Coordination control method based on vector machine online identification prediction |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114326406A (en) |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101038277A (en) * | 2007-04-19 | 2007-09-19 | 东北大学 | Soft measurement method for coal power fineness in powdering producer |
| CN104574220A (en) * | 2015-01-30 | 2015-04-29 | 国家电网公司 | Power customer credit assessment method based on least square support vector machine |
| CN109034223A (en) * | 2018-07-16 | 2018-12-18 | 湖南城市学院 | A kind of Study on prediction technology of chaotic series and system |
| CN110598900A (en) * | 2019-07-31 | 2019-12-20 | 成都三六八建设工程有限公司 | Electric power load prediction algorithm based on genetic algorithm and support vector machine |
| AU2020104000A4 (en) * | 2020-12-10 | 2021-02-18 | Guangxi University | Short-term Load Forecasting Method Based on TCN and IPSO-LSSVM Combined Model |
| CN113850320A (en) * | 2021-09-24 | 2021-12-28 | 沈阳理工大学 | Transformer fault detection method based on improved support vector machine regression algorithm |
-
2021
- 2021-12-31 CN CN202111669703.XA patent/CN114326406A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101038277A (en) * | 2007-04-19 | 2007-09-19 | 东北大学 | Soft measurement method for coal power fineness in powdering producer |
| CN104574220A (en) * | 2015-01-30 | 2015-04-29 | 国家电网公司 | Power customer credit assessment method based on least square support vector machine |
| CN109034223A (en) * | 2018-07-16 | 2018-12-18 | 湖南城市学院 | A kind of Study on prediction technology of chaotic series and system |
| CN110598900A (en) * | 2019-07-31 | 2019-12-20 | 成都三六八建设工程有限公司 | Electric power load prediction algorithm based on genetic algorithm and support vector machine |
| AU2020104000A4 (en) * | 2020-12-10 | 2021-02-18 | Guangxi University | Short-term Load Forecasting Method Based on TCN and IPSO-LSSVM Combined Model |
| CN113850320A (en) * | 2021-09-24 | 2021-12-28 | 沈阳理工大学 | Transformer fault detection method based on improved support vector machine regression algorithm |
Non-Patent Citations (3)
| Title |
|---|
| 沈利 等: "基于燃煤特性的电站锅炉排烟温度预测模型研究", 热力发电, no. 07, pages 19 - 23 * |
| 钟伟民 等: "一种基于2 次核函数支持向量机的多步预测控制算法", 华东理工大学学报(自然科学版), vol. 33, no. 4, pages 524 - 528 * |
| 陈果: "基于遗传算法的支持向量机时间序列预测模型优化", 仪器仪表学报, vol. 27, no. 9, pages 1080 - 1084 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108564204B (en) | Least square support vector machine electricity quantity prediction method based on maximum correlation entropy criterion | |
| CN110598929B (en) | Wind power nonparametric probability interval ultrashort term prediction method | |
| CN110989342B (en) | Real-time T-S fuzzy modeling method for combined cycle unit heavy-duty gas turbine | |
| CN108167802B (en) | Multi-model intelligent optimizing and predicting control method for boiler load under low load | |
| CN110009135B (en) | Wind power prediction method based on width learning | |
| CN112836884A (en) | Copula-DBiLSTM-based method for accurately predicting multi-element load of comprehensive energy system | |
| CN115374995A (en) | Distributed photovoltaic and small wind power station power prediction method | |
| CN110866633A (en) | Micro-grid ultra-short term load prediction method based on SVR support vector regression | |
| CN114219181A (en) | Wind power probability prediction method based on transfer learning | |
| CN111723982A (en) | Medium-and-long-term power load combined prediction method based on gray-Markov chain | |
| CN111832839B (en) | Energy consumption prediction method based on sufficient incremental learning | |
| CN112149883A (en) | Photovoltaic power prediction method based on FWA-BP neural network | |
| CN112052570A (en) | Economy backpressure optimization method of wet cooling unit of thermal power plant based on wolf algorithm | |
| CN115759415A (en) | Power consumption demand prediction method based on LSTM-SVR | |
| CN112668234A (en) | Intelligent control method for steelmaking endpoint of converter | |
| CN113591957B (en) | Wind power output short-term rolling prediction and correction method based on LSTM and Markov chain | |
| CN111192158A (en) | Transformer substation daily load curve similarity matching method based on deep learning | |
| CN116865343B (en) | Model-free self-adaptive control method, device and medium for distributed photovoltaic power distribution network | |
| CN114139439A (en) | Steam turbine optimal initial pressure determination method based on simulated annealing particle swarm algorithm | |
| CN110276478B (en) | Short-term wind power prediction method based on segmented ant colony algorithm optimization SVM | |
| CN114326406A (en) | Coordination control method based on vector machine online identification prediction | |
| CN110909492A (en) | Sewage treatment process soft measurement method based on extreme gradient lifting algorithm | |
| CN114091304B (en) | Intelligent decision-making method for processing hull plate by oxyhydrogen gas heat source | |
| CN116292057A (en) | Control method, device, equipment and medium for variable-speed pumped storage unit | |
| CN114861555A (en) | Regional comprehensive energy system short-term load prediction method based on Copula theory |
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 |