CN108536127B - A kind of model mismatch diagnostic method of the multivariable control system of data-driven - Google Patents
A kind of model mismatch diagnostic method of the multivariable control system of data-driven Download PDFInfo
- Publication number
- CN108536127B CN108536127B CN201810357053.7A CN201810357053A CN108536127B CN 108536127 B CN108536127 B CN 108536127B CN 201810357053 A CN201810357053 A CN 201810357053A CN 108536127 B CN108536127 B CN 108536127B
- Authority
- CN
- China
- Prior art keywords
- model
- control system
- multivariable control
- output
- mismatch
- 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
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
- G05B23/00—Testing or monitoring of control systems or parts thereof
- G05B23/02—Electric testing or monitoring
- G05B23/0205—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Automation & Control Theory (AREA)
- Feedback Control In General (AREA)
Abstract
The invention discloses a kind of model mismatch diagnostic methods of the multivariable control system of data-driven, including according to close loop maneuver data, estimate white noise, recognize interference model, obtain the process model residual error of each output channel of global process's model residual sum.World model's quality index is obtained according to white noise and global process's model residual error, the overall process model mismatch of multivariable control system is detected according to world model's quality index, the overall model quality index and local model quality index of output channel are obtained according to the process model residual error of white noise and each output channel, for determining the input and output subchannel of the model mismatch of multivariable control system, for the partial model quality index of current subchannel closer to 0, model mismatch existing for the process model of the input and output subchannel is more serious.The present invention is diagnosable process model mismatch under industrial process accidental conditions, reduces system maintenance cost, improves security of system.
Description
Technical field
The invention belongs to technical field of industrial control, more particularly, to a kind of multivariable control system of data-driven
Model mismatch diagnostic method.
Background technique
Since the controller of industrial control system is that control engineer designs according to process model, thus process model pair
It is studied in control performance optimization and fault diagnosis etc. extremely important.Researcher designs classical PID controller according to process model
Ratio, integral and differential parameter.The output of process of MPC controller use process model and historical data prediction future, thus
Generate optimal manipulating variable.After system puts into operation, the dynamic characteristic of controlled device is always constantly changing, to occur
Journey model mismatch.If process model and practical object mismatch are serious, the mismatch of process model necessarily affects the controlling of system
Can, it is unstable to even result in closed-loop control system.Thus, the model mismatch for detecting closed-loop control system becomes most important.
However, actual industrial process generally comprises multiple I/O channels, and there is coupling between channel.In recent years
Come, researcher both domestic and external is dedicated to having detected the model mismatch of multivariable control system.Some researchers are using intrusion
The method of formula has estimated the interference model of real system, utilizes the interference model of estimation and the time series mould of reality output error
Difference between type has detected the model mismatch of multivariable control system.Some researchers propose a kind of model quality and comment
Estimate technology, has rated the overall model quality of process model and interference model.These methods have detected multivariable Control closed loop control
World model's mismatch of system processed.When there is severe mismatch in process model, it is necessary to identification process model again.
In fact, the identification of process model is a quite time-consuming process, and recognize the process of multivariable control system
Model is more complicated.For example, one 5 × 5 multivariable control system includes 25 autonomous channels, it would be desirable to recognize 25 sons
Model could complete the identification of whole system, necessarily will cause waste economically.Then, researcher is using intrusive side
Method has deeply diagnosed the model mismatch of multivariable control system, it is determined that the subchannel or submodel of mismatch.In the prior art, have
Following several method is mismatched for detection model and object: (1) being analyzed between process input jiffer signal and prediction error
Correlativity, so that the submodel of mismatch is separated from the overall model of mimo systems;(2) model is calculated
Partial correlation coefficient between residual sum manipulating variable proposes a kind of side of mismatch subchannel for recognizing model predictive control system
Method;(3) difference between the Controlling model and practical object of raw ore grinding circuit is had detected;(4) when good according to control performance
Markov Parameters statistics band and control performance degenerate when statistics band deviation, diagnosed the subchannel of model mismatch;
(5) subspace projection method and assumed statistical inspection method are combined, it is determined that model mismatch in separate manufacturing firms model
Element;(6) a kind of method of two steps diagnosis is proposed, it is determined that the exact position of model mismatch in target transfer function matrix;
(7) CN105807611A is the immediate prior art of the application, and the model and object for describing a kind of closed-loop control system are not
Matched detection method, comprising: acquisition close loop maneuver data, and centralization processing is carried out to the close loop maneuver data;It is described
Close loop maneuver data include the output of process and the process input of closed-loop system;According to closed-loop control system structure, by just trading
Image method obtains interference and updates;According to reference signal and the output of process, the tracking error of closed-loop control system is obtained, and
Centralization processing is carried out to the tracking error;According to the interference update and the tracking error after centralization, by adaptive
Least absolute value is shunk and selection operator method, establishes the interference model of closed-loop control system;According to the close loop maneuver data
With the interference model of the closed-loop control system, the model quality variable of closed-loop control system is obtained;It is updated according to the interference
With the model quality variable, model and the unmatched index of object for detecting closed-loop control system, i.e. closed loop control are obtained
The model quality index of system processed;Closed-loop control system is detected using the model quality index according to closed-loop control system structure
Whether the model of system matches with object.However, the above method is required when separating the mismatch submodel of multivariable control system
It is middle on the set point or manipulating variable of the industrial process of normal operation that certain dynamic excitation is added, it necessarily will affect control system
The normal operation of system.
It is difficult to diagnose process model mismatch under industrial process accidental conditions, is it can be seen that the prior art exists
The technical problem that maintenance cost of uniting is high, security of system is low.
Summary of the invention
Aiming at the above defects or improvement requirements of the prior art, the present invention provides a kind of multivariable Controls of data-driven
The model mismatch diagnostic method of system, thus the solution prior art, which exists, is difficult to diagnose under industrial process accidental conditions
Journey model mismatch, the technical problem that system maintenance is at high cost, security of system is low.
To achieve the above object, the present invention provides a kind of diagnosis of the model mismatch of the multivariable control system of data-driven
Method, comprising:
(1) according to the close loop maneuver data of multivariable control system, the white noise estimated and output error, Jin Ertong
Adaptive least absolute value contraction and selection operator are crossed, the interference model of multivariable control system is obtained;
(2) according to the interference model of close loop maneuver data and multivariable control system, the complete of multivariable control system is obtained
The process model residual error of office process model residual error and each output channel, by the variance and multivariable control system of the white noise of estimation
Global process's model residual error variance world model quality index of the quotient as multivariable control system, utilize multivariable control
System processed world model's quality index detection multivariable control system overall process model whether mismatch;
(3) if the overall process model mismatch of multivariable control system, by the white noise of i-th of output channel of estimation
The quotient of the variance of the process model residual error of i-th of output channel of variance and multivariable control system is as multivariable control system
I-th of output channel overall model quality index, utilize the overall model quality index detection i-th of i-th output channel
The process model of a output channel whether mismatch;
(4) if the process model mismatch of i-th of output channel, according to the white noise and i-th of all output channels of estimation
The process model residual error of a output channel, is shunk and selection operator by adaptive least absolute value, and that is estimated is all defeated
Linear regression model (LRM) between the white noise in channel and the process model residual error of i-th of output channel out, according to linear regression mould
The process model residual error of the coefficient of type, the white noise of all output channels of estimation and i-th of output channel, obtain i-th it is defeated
Path partially model quality index out diagnoses multivariable control system using the partial model quality index of i-th of output channel
Each input and output subchannel process model whether mismatch.
Further, step (1) includes:
(1.1) the close loop maneuver data for acquiring multivariable control system carry out centralization processing, institute to close loop maneuver data
State the output of process and process input that close loop maneuver data include closed-loop system;
(1.2) according to the difference between the setting value and the output of process of process input, the defeated of multivariable control system is obtained
Error out, and centralization processing is carried out to output error;
(1.3) after to the setting value centralization processing of the output of process, white noise and center in conjunction with multivariable control system
Change treated the output of process, the autoregression model with external source import of establishment process output, and then passes through rectangular projection side
Method obtains the white noise of multivariable control system estimation;
(1.4) according to the white noise and centralization treated output error of multivariable control system estimation, by adaptive
Least absolute value contraction and selection operator are answered, the interference model of multivariable control system is obtained.
Further, step (1.4) includes:
According to the white noise and output error of multivariable control system, the line of the output error of multivariable control system is established
Property regression model, by the white noise of multivariable control system estimation and centralization, treated that output error brings output error into
Linear regression model (LRM), the ARMA model with external source import of the output error after obtaining centralization, by certainly
Least absolute value contraction and selection operator are adapted to, the autoregression with external source import of the output error after obtaining centralization is mobile
Averaging model coefficient vector, the transfer function matrix of the interference model for estimating multivariable control system.
Further, global process's model residual error of multivariable control system are as follows:
Wherein, ν (t) is global process's model residual error of multivariable control system,For multivariable control system
Inverse, the G of the transfer function matrix of the interference model of middle estimationmod(z-1) be multivariable control system process model transmitting letter
Number, y (t) are the output of process of multivariable control system, and u (t) is that the process of multivariable control system inputs.
Further, when world model's quality index of multivariable control system is less than 0.98, multivariable control system
Overall process model mismatch.
Further, when the overall model quality index of i-th of output channel of multivariable control system is less than 0.98, the
The process model mismatch of i output channel.
Further, the partial model quality index of i-th of output channel are as follows:
Wherein, t is t-th of sampling instant,For the variance of the white noise of i-th of output channel of estimation,For the variance of the white noise of j-th of input channel of estimation,For the order of the linear regression model (LRM) of estimation,For the coefficient of the linear regression model (LRM) of estimation, ηL,ijFor i-th of output channel of j-th of multivariable control system input
Partial model quality index,
Further, when the partial model quality index of i-th of output channel is less than 0.98, the input and output subchannel
Process model mismatch.
In general, through the invention it is contemplated above technical scheme is compared with the prior art, can obtain down and show
Beneficial effect:
(1) present invention is when diagnosing the process model mismatch of multivariable control system, according to process model residual error and estimation
White noise between relationship, the overall model quality index of world model's quality index and output channel and local model quality
The influence of the change of the adjustment parameter of the uncontrolled device of index;Relative to the index of other multivariable control systems, due to this hair
The regulatory factor of world model's quality index of bright proposition and two partial model quality index of output channel and controller
Change unrelated, method proposed by the present invention can effectively control the process model mismatch of multivariable control system from other influences
It is separated in the factor of performance, this provides effective foundation for the reason of diagnosis control performance degradation.
(2) diagnostic method of the model mismatch of the multivariable control system of data-driven provided by the invention can be assessed more
World model's quality of variable control system can effectively determine the input and output subchannel there are model mismatch again, work as process
When severe mismatch occurs in model, the process model of identification system is needed again, this method only needs identification, and there are the inputs of mismatch
Subchannel is exported, avoids bringing unnecessary economic loss to enterprise, has saved production cost.In addition, side provided by the invention
Method has diagnosed the model mismatch of multivariable control system using the method for non-intrusion type, under industrial process accidental conditions i.e.
Diagnosable process model mismatch, reduces system maintenance cost, improves security of system.
Detailed description of the invention
Fig. 1 is a kind of model mismatch diagnostic method of the multivariable control system of data-driven provided in an embodiment of the present invention
Flow chart;
Fig. 2 is variable control system structural schematic diagram provided in an embodiment of the present invention;
Fig. 3 is the schematic diagram for the Wood-Berry rectifying column process that the embodiment of the present invention 1 provides.
Specific embodiment
In order to make the objectives, technical solutions, and advantages of the present invention clearer, with reference to the accompanying drawings and embodiments, right
The present invention is further elaborated.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, and
It is not used in the restriction present invention.As long as in addition, technical characteristic involved in the various embodiments of the present invention described below
Not constituting a conflict with each other can be combined with each other.
As shown in Figure 1, a kind of diagnostic method of the model mismatch of the multivariable control system of data-driven, specific as follows:
(1) according to the close loop maneuver data of multivariable control system, the white noise estimated and output error, Jin Ertong
Adaptive least absolute value contraction and selection operator are crossed, the interference model of multivariable control system is obtained;
(2) according to the interference model of close loop maneuver data and multivariable control system, the complete of multivariable control system is obtained
The process model residual error of office process model residual error and each output channel, by the variance and multivariable control system of the white noise of estimation
Global process's model residual error variance world model quality index of the quotient as multivariable control system, utilize multivariable control
System processed world model's quality index detection multivariable control system overall process model whether mismatch;
(3) if the overall process model mismatch of multivariable control system, by the white noise of i-th of output channel of estimation
The quotient of the variance of the process model residual error of i-th of output channel of variance and multivariable control system is as multivariable control system
I-th of output channel overall model quality index, utilize the overall model quality index detection i-th of i-th output channel
The process model of a output channel whether mismatch;
(4) if the process model mismatch of i-th of output channel, according to the white noise and i-th of all output channels of estimation
The process model residual error of a output channel, is shunk and selection operator by adaptive least absolute value, and that is estimated is all defeated
Linear regression model (LRM) between the white noise in channel and the process model residual error of i-th of output channel out, according to linear regression mould
The process model residual error of the coefficient of type, the white noise of all output channels of estimation and i-th of output channel, obtain i-th it is defeated
Path partially model quality index out diagnoses multivariable control system using the partial model quality index of i-th of output channel
Each input and output subchannel process model whether mismatch.
Preferably, step (1) includes:
The mean vector of the process input of multivariable control system is obtained according to the following formula:
Wherein,For the mean vector of the process input tieed up of m × 1, m indicates that m input of multivariable control system is logical
Road, t indicate t-th of sampling instant, and N indicates sampled data number, u1(t) indicate that first input of multivariable control system is logical
The process in road inputs, um(t) the process input of m-th of input channel of multivariable control system is indicated;
The mean vector of the output of process of multivariable control system is obtained according to the following formula:
Wherein,For the mean vector of the output of process tieed up of n × 1, n indicates that n output of multivariable control system is logical
Road, t indicate t-th of sampling instant, and N indicates sampled data number, y1(t) indicate that first output of multivariable control system is logical
The output of process in road, yn(t) the output of process of n-th of output channel of multivariable control system is indicated;
Centralization processing is carried out to the process input of m input channel of multivariable control system according to the following formula:Wherein, u (t) indicates the process input that m × 1 is tieed up,Indicate the equal of the process input that m × 1 is tieed up
It is worth vector, u ' (t) indicates centralization treated process input.
Centralization processing is carried out to the output of process of n output channel of multivariable control system according to the following formula:Wherein, y (t) indicates the output of process that n × 1 is tieed up,Indicate the output of process that n × 1 is tieed up
Mean vector, y ' (t) indicate centralization treated the output of process.
The output error of multivariable control system: e (t)=y (t)-r (t), wherein e (t) indicates that the output that n × 1 is tieed up misses
Difference, y (t) indicate that the output of process that n × 1 is tieed up, r (t) indicate the reference signal for the output of process that n × 1 is tieed up.
The mean value of the output error of multivariable control system is obtained according to the following formula:
Wherein, t indicates t-th of sampling instant.For the mean vector of the output error tieed up of n × 1, n indicates multivariable
N output channel of control system, N indicate sampled data number, e1(t) indicate that first output of multivariable control system is logical
The output error in road, en(t) output error of n-th of output channel of multivariable control system is indicated;
Centralization processing is carried out to output error:Wherein, e (t) indicates the output that n × 1 is tieed up
Error, e ' (t) indicate centralization treated output error,For the mean value of the output error tieed up of n × 1.
Centralization processing is carried out to the setting value of the output of process of multivariable control system;According to multivariable control system
Structure, the autoregression model with external source import of establishment process output are specific as follows:
Wherein, M1Indicate the order of the autoregression item of the autoregression model with external source import, HiIt indicates to have external source import
Autoregression model autoregression item coefficient matrix, M2Indicate the external source import item of the autoregression model with external source import
Order, LiIndicate that the coefficient matrix of the external source import item of the autoregression model with external source import, t indicate t-th of sampling instant,
Y ' (t) indicates the output of process that centralization treated n × 1 is tieed up, and r ' (t) indicates centralization treated the setting that n × 1 is tieed up
Value, ε (t) are that the n × 1 of multivariable Control coefficient ties up white noise;
According to centralization treated the output of process y ' (t) and centralization treated setting value r ' (t), following number is established
According to matrix:
y′P(t)=[y ' (t) ..., y ' (t-P+1)], r 'P(t)=[r ' (t) ..., r ' (t-P+1)]
Wherein, t indicates t-th of sampling instant, and P indicates the data window size of estimation white noise, y 'P(t) center is indicated
Change n × P that treated the output of process y ' (t) is constituted and tie up matrix, y ' (t) indicates centralization treated the output of process, r ' (t)
Indicate centralization treated setting value, r 'P(t) indicate that n × P that centralization treated setting value r ' (t) is constituted ties up matrix,
XP(t) (nM is indicated1+nM2The data matrix of) × P dimension;
Multivariable control system is estimated by rectangular projection method according to the autoregression model with external source import
White noise:
Wherein, t indicates t-th of sampling instant, and I is the unit matrix of P × P dimension,Indicate the white noise of n × P dimension estimation
Sound, whereinIndicate the white noise of estimation.
The mean vector of the setting value of multivariable control system is obtained according to the following formula:
Wherein, t indicates t-th of sampling instant,For the mean vector of the setting value tieed up of n × 1, n indicates multivariable control
N output channel of system processed, N indicate sampled data number, r1(t) first output channel of multivariable control system is indicated
The output of process setting value, rn(t) setting value of the output of process of n-th of output channel of multivariable control system is indicated;
Centralization processing is carried out to setting value according to the following formula:Wherein, r (t) indicates that n × 1 is tieed up
Setting value, r ' (t) indicates centralization treated setting value,For the mean vector of the setting value tieed up of n × 1.
According to the closed loop configuration of multivariable control system, establishes the linear of the output error of entire multivariable control system and return
Return model, be expressed as follows:
E (t)=- T (z-1)e(t-1)-r(t)+Gd(z-1)ε(t)
Wherein, z-1 indicates delay factor, T (z-1) indicate the transfer function matrix of stringent cause and effect, Gd(z-1) indicate multivariable control
The interference model transfer function matrix of system processed, e (t) indicate output error, and r (t) indicates setting value, and ε (t) indicates white noise.
According to the linear regression model (LRM) of the output error, n output channel of multivariable control system is successively established
The linear regression model (LRM) of output error:
Wherein, t indicates t-th of sampling instant, z-1Indicate delay factor, e1(t) output of first output channel is indicated
Error, en(t) output error of n-th of output channel of multivariable control system is indicated, n indicates the of multivariable control system
N output channel, T11(z-1) ..., T1n(z-1) ..., Tn1(z-1) ..., Tnn(z-1) indicate the transmission function of stringent cause and effect,Indicate the transmission function of the interference model of first output channel,Indicate the defeated of n-th of output channel
Error out, r1(t) setting value of the output of process of first output channel of multivariable control system, r are indicatedn(t) indicate changeable
The setting value of the output of process of n-th of output channel of amount control system, ε1(t) white noise of first output channel, ε are indicatedn
(t) white noise of n-th of output channel is indicated;
According to the linear regression model (LRM) of the output error of n output channel of the multivariable control system and it is adaptive most
Small absolute value is shunk and the interference model of selection operator method estimation n output channel of multivariable control system, specifically: according to
The linear regression model (LRM) of the output error of n output channel of the multivariable control system establishes first output channel
The ARMA model with external source import of output error after centralization, is expressed as follows:
Wherein, t indicates t-th of sampling instant, z-1Indicate delay factor,Indicate first output channel
Output error autoregression part coefficient, Indicate the external source import of the output error of first output channel
Partial coefficient, whereinIndicate first output channel
Output error rolling average part coefficient, I1Indicate the autoregression part of the output error of first output channel most
Big order, I12..., I1nAnd J1Indicate the maximum order of the external source import part of the output error of first output channel, K1Table
Show the maximum order of the rolling average part of the output error of first output channel, e '1(t) first output channel is indicated
Output error after centralization, r '1(t) setting value after indicating the centralization of first output channel,Indicate estimation
The white noise of first output channel;
It is missed according to the output of the white noise of first output channel and centralization treated first output channel
Difference constructs following data matrix:
E1,W(t)=[e '1(t) … e′1(t-W+1)]T
Wherein, W indicates the window size of sampled data, E1,W(t) the dimension data matrix of W × 1, H are indicated1,W(t) indicate W ×
(I1+I12+…+I1n+J1+K1) dimension data matrix;
It is shunk according to adaptive least absolute value and selection operator, acquisition has the auto regressive moving average mould of external source import
Type coefficient vector:
Wherein, t indicates t-th of sampling instant,Indicate the band that adaptive least absolute value is shunk and selection operator is estimated
There is the coefficient vector of the ARMA model of external source import, i.e.,Above-mentioned for estimation has external source
The coefficient of the ARMA model of input, Τ indicate the practical but unknown auto regressive moving average with external source import
The coefficient vector of type, whereinλ indicates adaptive least absolute value contraction and selection operator
The adjustable parameter of penalty factor is controlled in method,Indicate j-th of coefficient φ in the coefficient of ARMA modeljPower
Repeated factor;
According to the linear regression model (LRM) of the output error of first output channel, the interference model of closed-loop system is obtained,
Discrete transfer function indicates are as follows:
Wherein,For estimation
The ARMA model with external source import coefficient,For the interference of first output channel of estimation
The transmission function of model.Successively estimate the interference model of remaining n-1 output channel of multivariable control system.
Preferably, step (2) includes:
According to the closed-loop control system structure of multivariable control system, global process's model of multivariable control system is obtained
Residual error indicates are as follows:
Wherein, ν (t) is global process's model residual error of multivariable control system,For above-mentioned multivariable Control
The transfer function matrix for the interference model estimated in system it is inverse, d be interference, Gmod(z-1) it is above-mentioned multivariable control system
The transmission function of process model.
According to the white noise of global process the model residual error and the multivariable control system of the multivariable control system,
Obtain world model's quality index of multivariable control system:
Wherein,Indicate the variance of the white noise of estimation, < v (t)2The global mistake of > expression multivariable control system
The variance of journey model residual error, ηGIndicate world model's quality index of multivariable control system, T indicates transposition.
According to the closed loop configuration of multivariable control system, the closed loop expression formula of global process's model residual error is obtained:
Wherein, t indicates t-th of sampling instant, z-1Indicate that delay factor, v (t) indicate the global mistake of multivariable control system
Journey model residual error, r (t) indicate setting value, and ε (t) indicates that the white noise of t-th of sampling instant, ε (t-1) indicate the t-1 sampling
The white noise at moment,Indicate inverse, the G of the interference model transfer function matrix of the multivariable control system of estimationd(z-1)
Indicate the interference model transfer function matrix of actual multivariable control system, Δ G (z-1) indicate multivariable control system reality
Target transfer function Gp(z-1) the process model transmission function G that uses with controllermod(z-1) difference, i.e. Δ G (z-1)=Gp(z-1)-Gmod(z-1),Indicate the transmission function of cause and effect, whereinQ(z-1) indicate more
The transfer function matrix of the controller of variable control system, I indicate unit matrix;
When setting value r (t) is when constant reference signal, the closed loop expression formula of above-mentioned global process's model residual error can be with
Simplify are as follows:
Above-mentioned expression formula shows that global process model residual error v (t) is a moving average process, and wherein white noise ε (t) is
The driving noise of the moving average process;
According to the relationship of global process model residual error ν (t) and white noise ε (t), world model's quality index η is obtainedGTake
It is worth range, ηG∈(0,1];
If world model quality index ηGClose to 1 i.e. range in 0.98-1, then the process mould of multivariable control system
Type and practical object exactly match;, whereas if world model quality index ηGLess than 0.98, then illustrate that entire closed-loop system is deposited
In process model mismatch;
Preferably, step (3) includes:
According to the structure of multivariable control system, the process model residual error of global process's model residual error and each output channel
Relationship is:
V (t)=[v1(t) … vi(t) … vn(t)]T
Wherein, t indicates t-th of sampling instant, v1(t) the process model residual error of first output channel, v are indicatedi(t) table
Show the process model residual error of i-th of output channel, vn(t) the process model residual error of n-th of output channel is indicated, v (t) indicates complete
Office's process model residual error;
According to global process's model residual error, the process model residual error of i-th of output channel: v can be obtainedi(t)=v (t)
(i);
I-th of the output according to the process model residual error of i-th of output channel and the multivariable control system is logical
The white noise in road, obtains the overall model quality index of i-th of output channel, i.e., the overall model quality of single output channel refers to
Mark:
Wherein, t indicates t-th of sampling instant,Indicate the white noise of i-th of output channel of estimation, vi(t) it indicates
The process model residual error of i-th of output channel,Indicate white noiseVariance, < vi(t)2I-th of > expression defeated
The process model residual error v in channel outi(t) variance, ηL,iIndicate the partial model quality index of i-th of output channel.
According to global process's model residual error and the relationship of white noise it is found that if ηL,iClose to 1, then show < vi(t)2>≈
<εi(t)2>, i.e. vi(t)≈εi(t), so that process model mismatch is not present in the process model of i-th of output channel;If ηL,i
Very little is even close to 0, then illustrating the process model of i-th of output channel, there are serious model mismatches.
As the partial model quality index η of i-th of output channelL,iWhen less than 0.98, (4) are thened follow the steps.
Preferably, step (4) includes:
According to the closed loop configuration of multivariable control system, the process model residual error and each output of above-mentioned i-th of output channel
The relationship of the white noise in channel establishes the process model residual error of i-th of output channel and having for the white noise of each output channel
The ARMA model of external source import, is expressed as follows:
Wherein, t indicates t-th of sampling instant, z-1Indicate delay factor, vi(t) the process mould of i-th of output channel is indicated
Type residual error, εi(t) white noise of i-th of output channel, i indicate i-th of output channel, εj(t) j-th of output channel is indicated
White noise,The rolling average part of ARMA model of the expression with external source import is
Number,Indicate the coefficient of the external source import part of the ARMA model with external source import, Ni
Indicate the maximum order of the rolling average part of the ARMA model with external source import, Nj(j=1,2 ..., n;j
≠ i) it indicates to have the maximum order of the external source import part of the ARMA model of external source import;
According to the process model residual error of the white noise of i-th of output channel and i-th of output channel, following number is constructed
According to matrix:
Vi,W(t)=[vi(t) … vi(t-W+1)]T
Wherein, W indicates the window size of sampled data, Vi,W(t) the dimension data matrix of W × 1, Σ are indicatedi,W(t) indicate W ×
(N1+…+Nn) dimension data matrix;
It is shunk according to adaptive least absolute value and selection operator method, autoregression of the acquisition with external source import is mobile flat
The coefficient vector and order vector of equal model:
Wherein, t indicates t-th of sampling instant,Indicate that adaptive least absolute value is shunk and selection operator method is estimated
The ARMA model with external source import coefficient vector, i.e.,Indicate that adaptive least absolute value is shunk and selection operator method
The order vector of the ARMA model with external source import of estimation, fkFor the returning certainly with external source import of estimation
Return the coefficient of moving average model(MA model), f indicate the coefficient of the practical but unknown auto regressive moving average type with external source import to
Amount, γ indicate the adjustable parameter that adaptive least absolute value shrinks and controls penalty factor in selection operator method,It indicates certainly
Return k-th of coefficient f in the coefficient of moving average model(MA model)kWeight factor;
According to the coefficient vector of estimation and order vector, white noise and the single output for obtaining i-th of output channel are logical
Linear regression model (LRM) between the process model residual error in road.
According to the mistake of the coefficient of linear regression model (LRM), the white noise of single output channel and i-th of output channel
Journey model residual error obtains the partial model quality index in single-input single-output channel:
Wherein, t indicates t-th of sampling instant,Indicate the white noise variance of i-th of output channel of estimation,Indicate the white noise variance of j-th of output channel of estimation,Indicate the white of i-th of output channel of estimation
Noise,Indicate the white noise of j-th of output channel of estimation,Indicate the order of the linear regression model (LRM) of estimation,Indicate the coefficient of the linear regression model (LRM) of estimation, ηL,ijIndicate that j-th of input, i-th of the output of multivariable control system is logical
The partial model quality index in road.
Single according to linear regression model (LRM), the process model residual error of single output channel of the estimation and the estimation exports
The variance of the white noise in channel, the process model residual error of i-th of output channel indicates are as follows:
Wherein, t indicates t-th of sampling instant, < vi(t)2The variance of the process model residual error of i-th of output channel of > expression,Indicate the white noise variance of i-th of output channel,Indicate the white noise of j-th of output channel of estimation
Variance,WithIndicate the order of the linear regression model (LRM) of estimation,WithThe linear regression model (LRM) for indicating estimation is
Number, n indicate the number of the output channel of multivariable control system;
According to the relationship of the white noise variance of the variance of the process model residual error of i-th of output channel and single output channel,
Obtain the partial model quality index η of i-th of output channelL,ijValue range;
If the partial model quality index η of i-th of output channelL,ijClose to 1, then illustrate that this input and output is logical
The process model and practical object in road are exact matchings;On the contrary, if ηL,ijLess than 0.98, then illustrate this input and output
The process model quality of subchannel, i.e. the process model mismatch of multivariable control system occur on this input and output subchannel.
Shown in Fig. 2, u (t), y (t) respectively indicate the process input of multivariable control system, the output of process with r (t) and set
Definite value, ε (t) indicate to close the white noise of multivariable control system, Q (z-1) indicate multivariable control system internal mode controller biography
Delivery function battle array, Gp(z-1) indicate multivariable control system practical object transfer function matrix, Gmod(z-1) indicate process model
Transfer function matrix, z-1Indicate delay factor;According to the closed loop configuration figure of multivariable control system, general multivariable Control system is obtained
The controller of system and the relationship of internal mode controller, specific as follows:
Gc(z-1)=[I-Q (z-1)Gmod(z-1)]-1Q(z-1);
Wherein, Gc(z-1) indicate multivariable control system in controller transfer function matrix, Gc(z-1) can be based on mould
The PID controller or MPC controller of type, I indicate unit matrix, Q (z-1) indicate the internal mode controller of multivariable control system
Transfer function matrix, Gmod(z-1) indicate process model transfer function matrix.
Embodiment 1
Wood-Berry rectifying column process as shown in figure 3, it be methanol-water mixing analogue simulation process, it be by
Wood and Berry put forward one is used to study the chemical process model with evaluation procedure control technology and performance monitoring.
The distillation process of Wood-Berry rectifying column it is specific as follows: (1) raw material D enters from feedboard, and feedboard is by entire rectifying
Tower is divided into rectifying section and stripping section;(2) after solution enters rectifying column, because first alcohol and water has different boiling points, so as to cause low
Boiling point component gasification is easier to and up ascension, and high boiling component then mostly with liquid to flow down, and with rise in rectifying column
Steam come into full contact on each layer column plate, to be conducted heat, mass transport process;(3) the liquid tower on earth flowed to raw material F
After kettle, partially returned in rectifying column again after the heated vaporization in part by continuous extraction to become bottom product;(4)
Upflowing vapor in rectifying column successively passes through whole column plates, so that the concentration of the volatile components in steam is gradually increased, on
The steam for rising to tower top is cooled to liquid in condenser, after reflux pump and return tank, is partially continuously drawn
To become overhead product, part is then introduced back as coolant liquid in tower to top tray, this process is exactly to flow back.
Wood-Berry rectifying column process includes two controlled variables (i.e. the output of process) and two manipulating variable (i.e. processes
Input), process transfer function matrix is as follows:
Wherein, Gp(s) the continuous transmission function of Wood-Berry rectifying column process is indicated, s indicates Laplace operator, behaviour
Vertical variable is return flow R (lb/min) and steam flow S (lb/min) respectively, and controlled variable is the top of the distillation column product respectively
Ingredient yT(mol%) and the ingredient y of bottom productB(mol%), raw material D regards the external interference on controlled variable as.
It is sampled by the data to Wood-Berry rectifying column process, the sampling time is 1 minute, available discrete
Transfer function matrix G after changep(z-1).Raw material D is nonstationary interference, and the transfer function matrix of interference model is:
Wherein, D (t) indicates nonstationary interference, Gd(z-1) indicate interference model transfer function matrix, z-1Expression is prolonged
The slow factor, θ1And θ2The parameter of interference model in the interference of expression process, in embodiment, θ1It is 0.95, θ2It is 0.7;ε (t) is
The white Gaussian noise that mean value is zero, it be mean value is zero that the covariance matrix of ε (t), which is white noise, and covariance matrix isThe setting value of the output of process of Wood-Berry rectifying column process is r (t)=[90mol%
5mol%].
The overhead product ingredient and bottom product of Wood-Berry rectifying column process are adjusted using model predictive controller
Ingredient.Model predictive controller (MPC) is a System design based on model device, its use process model generates prediction output, so
Optimal manipulating variable is obtained by minimizing the quadratic sum of prediction output error afterwards, to obtain the optimal behaviour at current time
Vertical variable.In embodiment, the continuous transmission function of Wood-Berry process model is obtained first, specific as follows:
Wherein, Gmod(s) the continuous transmission function of Wood-Berry process, K are indicatedm,11、Km,12、Km,21And Km,22It indicated
Journey model Gmod(s) gain, Tm,11、Tm,12、Tm,21And Tm,22Indicate TmIndicate process model Gmod(s) time constant,
τm,11、τm,12、τm,21And τm,22Indicate model Gmod(s) time delay.Secondly, " cp2dp " function according to the tool box MATLAB obtains
Take the discrete transfer function battle array G of Wood-berry rectifying column processmod(z-1).Then, the MPC tool case for selecting MATLAB to provide
To realize the emulation of multivariable closed-loop control system.In embodiment, the adjustment parameter of MPC controller is respectively: Hp、Hu、QyWith
Qu, wherein HpAnd HuRespectively indicate prediction time domain and control time domain, QyAnd QuRespectively indicate weight matrix and the manipulation of output error
The weight matrix of Delta.In embodiment, time domain H is predictedpIt is 100, controls time domain HuIt is 10, output error weight matrix Qy
It is unit matrix, the weight matrix Q of manipulating variable incrementuIt is null matrix.
The control performance of above-mentioned Wood-Berry rectifying column process is evaluated using Key Performance Indicator, Key Performance Indicator
Calculation formula is as follows:
Wherein, N indicates sampled data number, and t indicates t-th of sampling instant, and Δ u (t) indicates multivariable control system
The increment of manipulating variable, y (t) indicate that the output of process of multivariable control system, r (t) indicate the setting of multivariable control system
Value, QyIndicate output error weight matrix, QuIndicate the weight matrix of manipulating variable increment,Two norms of representing matrix,
The Key Performance Indicator of KPI expression multivariable control system.
Using the multivariable control system of data-driven provided by the invention model mismatch diagnostic method to embodiment
Wood-Berry rectifying column process carry out model mismatch diagnosis method, it is specific as follows:
(1) the close loop maneuver data for acquiring multivariable control system obtain the input of centralization treated process and process
Transmission of data
When multivariable control system operates normally, the sampling time is set as 1 minute, acquires closed-loop data, the sample of acquisition
This number N is 2000;According to the 2000 groups of process input datas collected, first the process input of acquisition multivariable control system
Mean value, calculation formula is:Then, after obtaining centralization
2000 groups of process input datas, calculation formula is:According to the 2000 of acquisition groups of the output of process numbers
According to, first acquisition multivariable control system the output of process mean value, calculation formula is:Then, 2000 groups of the output of process after centralization are obtained
Data, calculation formula are:
(2) output error of multivariable control system is obtained, and centralization processing is carried out to output error
Firstly, the output for obtaining multivariable control system misses according to the 2000 of acquisition groups of the output of process data and setting values
Difference, calculation formula are: e (t)=y (t)-r (t), then, obtain the mean value of the output error of multivariable control system, calculate public
Formula is:Centralization processing finally is carried out to output error,
Calculation formula is:
(3) white noise of multivariable control system is obtained
Firstly, obtaining the mean value of the setting value of multivariable control system, calculation formula according to the 2000 of acquisition groups of setting values
It is:Centralization treated setting value is being obtained, is being calculated
Formula is:
Secondly, the autoregression model with external source import of establishment process output, indicates are as follows:Wherein M1It is 50, M2It is 50;
Then, the centralization obtained according to step (1) and step (2) treated the output of process and setting value, is established such as
Lower data matrix:
y′P(t)=[y ' (t) ..., y ' (t-P+1)]
r′P(t)=[r ' (t) ..., r ' (t-P+1)]
Wherein, 1950 P, t 2000, M1It is 50, M2It is 50;
Finally, obtaining multivariable control system by rectangular projection method according to the data matrix that step (2) are established
White noise, calculation formula are:
(4) interference model of multivariable control system is obtained
Firstly, establishing the vector form of the linear regression model (LRM) of the output error of multivariable control system, it is expressed as follows:
E (t)=- T (z-1)e(t-1)-r(t)+Gd(z-1)ε(t);
Since Wood-Berry rectifying column process includes two outputs, the output error of multivariable control system
Vector form is indicated with two individual linear regression model (LRM)s, it may be assumed that
Then, estimate the interference model of first output channel:
(4.1) centralization is established treated the autoregression with external source import of the output error of first output channel
Moving average model(MA model) is expressed as follows:
(4.2) according to the defeated of the white noise of first output channel and centralization treated first output channel
Error out constructs following data matrix:
E1,W(t)=[e '1(t) … e′1(t-W+1)]T
Wherein, W indicates the window size of sampled data, E1,W(t) the dimension data matrix of W × 1, H are indicated1,W(t) W × (I is indicated
+ J) dimension data matrix;Wherein, t=1950, I1=20, J1=20, I2=20K1=20, W=1930;
(4.3) using the contraction of adaptive least absolute value and selection operator method, ARMA model coefficient vector is obtained:Wherein, t indicates t-th of sampling instant,It indicates
Adaptive least absolute value is shunk and the autoregression with external source import of the above-mentioned steps (4.4) of selection operator method estimation moves
The coefficient vector of dynamic averaging model, i.e.,For
The coefficient of the above-mentioned ARMA model with external source import of estimation, Τ indicate that practical but unknown to have external source defeated
The coefficient vector of the auto regressive moving average type entered, whereinλ indicates adaptive minimum absolute
Value shrinks the adjustable parameter with control penalty factor in selection operator method,Indicate the coefficient of ARMA model
In j-th of coefficient φjWeight factor.Wherein, best adjustable parameter λ is obtained by ten times of cross validation method, should
Method is shunk in adaptive least absolute value to be configured with inside selection operator method;It indicates with outer
The initial estimate of j-th of coefficient of ARMA model of source input,It is obtained using least square method, i.e.,
(4.4) formula is utilizedAcquisition first is defeated
The transmission function of the interference model in channel out, whereinWithTable
Show the coefficient of the ARMA model with external source import of step (4.3) estimation;
Finally, estimating Wood-Berry rectifying column according to step (4.1), step (4.2), step (4.3) and step (4.4)
The interference model of second output channel of process
(5) global process's model residual error of multivariable control system and the process model residual error of each output channel are obtained
Utilize formulaObtain the overall situation of multivariable control system
Process model residual error, wherein global process model residual error of the ν (t) for multivariable control system, t=1 ..., 2000;
(6) world model's quality index of multivariable control system is obtained
Utilize formulaObtain world model's quality of multivariable control system
Index ηG, wherein N=2000:
(7) detect multivariable control system overall process model whether mismatch
Using the closed loop configuration of multivariable control system, world model's quality index η is obtainedGValue range: ηG∈(0,
1];Again by world model quality index ηGIt is compared with 1, if world model's quality index is approximately equal to 1, shows multivariable
The process model of control system be it is optimal, that is, be not present model mismatch;Anyway, then the process model of multivariable control system is deposited
In mismatch, then the diagnosis for carrying out subsequent Sub-channel mode mismatch is needed;
(8) the overall model quality index of single output channel of multivariable control system is obtained
According to the structure of Wood-Berry rectifying column process, the mistake of global process's model residual error and two output channels is obtained
The relationship of journey model residual error, i.e. [v1(t) v2(t)]T=v (t);
The process model residual error v of two output channels is obtained again1(t) and v2(t), wherein v1(t) first output is indicated
The process model residual error in channel, v2(t) the process model residual error of second output channel of expression, t=1 ..., 2000;
According to formulaObtain the overall model quality index of two output channels;
(9) detect two output channels overall process model whether mismatch
According to the relationship of the white noise of the process model residual error of single output channel and single output channel it is found that if ηL,i(i
=1,2) close to 1, then show < vi(t)2>≈<εi(t)2> (i=1,2), i.e. vi(t)≈εi(t) (i=1,2), thus the i-th (i
=1,2) model mismatch is not present in the process model of a output channel;If ηL,i(i=1,2) very little is then said even close to 0
There are serious model mismatches for the process model of bright i-th (i=1,2) a output channel;When i-th (i=1,2) a output channel
There are model mismatches there are which input and output subchannel when mismatch, then needed to continue to diagnose for process model;
(10) the linear regression mould between the white noise of single output channel and the process model residual error of single output channel is recognized
Type
According to following formula:
Establish having between the process model residual error of i-th (i=1,2) a output channel and the white noise of each output channel
The ARMA model of external source import;
According to the process model residual error of the white noise of described two output channels and two output channels, following data are constructed
Matrix:
Vi,W(t)=[vi(t) … vi(t-W+1)]T, (i=1,2)
, (i=1,2), wherein W indicates the window size of sampled data, Vi,W(t), (i=1,2) indicates the dimension data of W × 1
Matrix, Σi,W(t), (i=1,2) indicates W × (N1+N2) dimension data matrix;
It is shunk according to adaptive least absolute value and selection operator method, the white noise for obtaining single output channel is exported with single
The coefficient vector and order vector of the ARMA model with external source import between the process model residual error in channel:Wherein,Indicate adaptive minimum
Absolute value is shunk and the coefficient vector of the above-mentioned ARMA model with external source import of selection operator method estimation,
I.e.Indicate that adaptive least absolute value is shunk and selection operator method is estimated
The order vector of the ARMA model with external source import of meter, fkFor estimation it is above-mentioned with external source import from
The coefficient of moving average model(MA model) is returned, f indicates the practical but unknown above-mentioned auto regressive moving average type with external source import
Coefficient vector, γ indicate the adjustable parameter that adaptive least absolute value shrinks and controls penalty factor in selection operator method,
Indicate k-th of coefficient f in the coefficient of ARMA modelkWeight factor;Wherein, best adjustable parameter γ is to pass through
What ten times of cross validation method obtained, this method is shunk in adaptive least absolute value to be set with inside selection operator method
It sets;Indicate the initial estimate of k-th of coefficient of ARMA model with external source import,It is obtained using least square method, i.e.,
Between the white noise of single output channel obtained according to above-mentioned steps and the process model residual error of single output channel
The coefficient vector and order vector of ARMA model with external source import obtain i-th of output channel
Linear regression model (LRM) between white noise and the process model residual error of i-th of output channel;
(11) the partial model quality index in single-input single-output channel is obtained
According to formulaObtain single-input single-output channel
Partial model quality index, wherein ηL,ijIndicate the localized mode of j-th of input, i-th of output channel of multivariable control system
Type quality index,Indicate the order of the linear regression model (LRM) of step (10) estimation,Indicate the linear of step (10) estimation
The coefficient of regression model;
(12) diagnose multivariable control system each input and output subchannel process model whether mismatch
According to formula
Obtain the value range of the partial model quality index in single-input single-output channel: ηL,ij∈ (0,1], (i=1,2;J=1,2);
If the partial model quality index η of i-th (i=1,2) a output channelL,ij(i=1,2) then illustrates this input close to 1
The process model and practical object for exporting subchannel are exact matchings;On the contrary, if ηL,ij(i=1,2) close to 0, then
Illustrate the process model quality of this input and output subchannel, i.e. it is defeated that this occurs for the process model mismatch of multivariable control system
Enter to export on subchannel;
(13) performance indicator of multivariable control system is obtained
Using formula Δ u (t)=u (t)-u (t-1), process input (the i.e. manipulation change of multivariable control system is first obtained
Amount) increment;Recycle formulaObtain multivariable Control system
The Key Performance Indicator of system, wherein N=2000.
According to the performance of Key Performance Indicator evaluation system, if performance indicator η0Smaller, the performance of system is poorer, if
Performance indicator η0Closer to 1, the performance of system is better.
In embodiment, we consider two different gain mismatch situations first.In the first case, gain mismatch
It is only present in R → yTIn channel, the process model in excess-three channel is all matched, i.e. Δ G11(q-1)=0.3Gp,11(q-1)
With Δ G12(q-1)=Δ G21(q-1)=Δ G22(q-1)=0.In second situation, channel R → yTWith channel S → yBMiddle difference
There is 30% and 20% gain mismatch, i.e. Δ G11(q-1)=0.3Gp,11(q-1), Δ G22(q-1)=0.2Gp,22(q-1) and Δ
G12(q-1)=Δ G21(q-1)=0.The model mismatch testing result of both of these case is given in table 1, wherein i is indicated i-th
Output channel.
1 channel R → y of tableTWith channel R → yBModel mismatch diagnosis when gain mismatch
In table 1, we calculate ηG、ηL,i(i=1,2) and ηL,ij(i=1,2;J=1,2).In the first case,
ηGIt is 0.79, this shows closed-loop control system, and there are process model mismatches.In addition, channel R → yTPartial model quality index be
0.75, i.e. ηL,11=0.75, and the overall model quality index of other subchannels is 1 or close to 1.According to the part of calculating
Model quality index, it is understood that process model mismatch appears in channel R → yT.In second situation, ηL,11=0.75,
ηL,22=0.80, i.e. channel R → yTPartial model quality index than channel S → yBPartial model quality index it is big.This
The result shows that channel R → yTWith channel S → yBIn all there is process model mismatch, and channel R → yTModel mismatch degree than logical
Road S → yBWant small.Therefore, the method that this chapter is proposed can not only diagnose process model mismatch, and can identify containing larger
The channel of smaller gain mismatch.
Secondly, diagnosing the process model mismatch of multivariable Control performance when the time delay of process model is there are when mismatch.Implement
In example, it is contemplated that two different process model Gm(s) the case where delay mismatch.In situation 1, delay mismatch appears in logical
Road R → yTIn, i.e.,
In situation 2, channel S → yTWith R → yBIn all there is delay mismatch, i.e.,
Table 2 gives the model mismatch testing result of both situations.
Model mismatch of 2 process model of table with delay mismatch diagnoses
In table 2, the η of situation 1GIt is 0.13, this shows that the process model of closed-loop control system and practical object seriously lose
Match.KPI at this time is 3.47, and Control performance standard when it is than model mismatch is much bigger.In addition, channel R → yTPart
Model quality index ηL,11It is 0.13, the partial model quality index of rest channels is 1 or close to 1, i.e. ηL,12=0.99 He
ηL,21=ηL,22=1, this indicates that process model mismatch occurs in channel R → yTIn.
In situation 2, ηL,12And ηL,21It is 0.53 and 0.37 respectively, and the partial model quality index of other subchannels connects
It is bordering on 1, i.e. ηL,11=ηL,22=0.98.According to the model mismatch diagnostic result of situation 2, we obtain S → yTWith R → yBTwo
All there is model mismatch in subchannel.Therefore, this method can also diagnose the delay mismatch of process model.
Finally, we consider controllers to adjust the influence to overall model quality index.In embodiment, three kinds are used
The adjustment parameter of different MPC controllers.The parameter setting of the first controller is as the controller parameter of basic condition, and second
The weight matrix of the manipulated variable increment of kind controller becomes Qu=diag { 0.2,0.3 }.In the third MPC controller, we
It joined constraint, i.e. -1mol%≤R≤5mol% on this performance variable of return flow R.As channel R → yTThere are processes
When model mismatch, i.e. Δ G11(q-1)=- 0.5Gp,11(q-1), the model mismatch detection of three kinds of different controllers is given in table 3
As a result.
Influence of the controller parameter change to model-evaluation index when 3 process model mismatch of table
As known from Table 3, when controller parameter changes, Control performance standard KPI changes between 0.62 and 0.80, and this
The overall model quality index that chapter is proposed all remains unchanged.The result of table 3 illustrates, when there are process model mistakes in process model
Timing, the change of MPC controller parameter has a significant impact to control performance, and world model's quality index and partial model matter
Figureofmerit is insensitive to the adjusting of controller.
As it will be easily appreciated by one skilled in the art that the foregoing is merely illustrative of the preferred embodiments of the present invention, not to
The limitation present invention, any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should all include
Within protection scope of the present invention.
Claims (8)
1. a kind of model mismatch diagnostic method of the multivariable control system of data-driven characterized by comprising
(1) according to the close loop maneuver data of multivariable control system, the white noise estimated and output error, and then by certainly
Least absolute value contraction and selection operator are adapted to, the interference model of multivariable control system is obtained;
(2) according to the interference model of close loop maneuver data and multivariable control system, the global mistake of multivariable control system is obtained
The process model residual error of journey model residual error and each output channel, by the complete of the variance of the white noise of estimation and multivariable control system
World model quality index of the quotient of the variance of office's process model residual error as multivariable control system, utilizes multivariable Control system
System world model's quality index detection multivariable control system overall process model whether mismatch;
(3) if the overall process model mismatch of multivariable control system, by the variance of the white noise of i-th of output channel of estimation
With the quotient of the variance of the process model residual error of i-th of output channel of multivariable control system as multivariable control system
The overall model quality index of i output channel, it is defeated using i-th of the detection of overall model quality index of i-th of output channel
Out the process model in channel whether mismatch;
(4) defeated according to the white noise of all output channels of estimation and i-th if the process model mismatch of i-th of output channel
The process model residual error in channel out, is shunk by adaptive least absolute value and selection operator, all outputs estimated are logical
Linear regression model (LRM) between the white noise in road and the process model residual error of i-th of output channel, according to linear regression model (LRM)
Coefficient, estimation all output channels white noise and i-th of output channel process model residual error, it is logical to obtain i-th of output
Road partial model quality index diagnoses each of multivariable control system using the partial model quality index of i-th of output channel
The process model of a input and output subchannel whether mismatch.
2. a kind of model mismatch diagnostic method of the multivariable control system of data-driven as described in claim 1, feature
It is, the step (1) includes:
(1.1) the close loop maneuver data for acquiring multivariable control system carry out centralization processing to close loop maneuver data, described to close
Ring operation data includes the output of process and the process input of closed-loop system;
(1.2) according to the difference between the setting value and the output of process of process input, the output for obtaining multivariable control system is missed
Difference, and centralization processing is carried out to output error;
(1.3) after to the setting value centralization processing of the output of process, at the white noise and centralization in conjunction with multivariable control system
The output of process after reason, the autoregression model with external source import of establishment process output, and then by rectangular projection method, it obtains
The white noise estimated to multivariable control system;
(1.4) according to the white noise and centralization treated output error of multivariable control system estimation, by adaptively most
Small absolute value is shunk and selection operator, obtains the interference model of multivariable control system.
3. a kind of model mismatch diagnostic method of the multivariable control system of data-driven as claimed in claim 2, feature
It is, the step (1.4) includes:
According to the white noise and output error of multivariable control system, establishes the linear of the output error of multivariable control system and return
Return model, treated that output error brings the linear of output error by white noise and the centralization of multivariable control system estimation
Regression model, the ARMA model with external source import of the output error after obtaining centralization, by adaptive
Least absolute value is shunk and selection operator, the auto regressive moving average with external source import of the output error after obtaining centralization
Model coefficient vector, the transfer function matrix of the interference model for estimating multivariable control system.
4. a kind of model mismatch diagnostic method of the multivariable control system of data-driven as claimed in claim 1 or 2, special
Sign is, global process's model residual error of the multivariable control system are as follows:
Wherein, v (t) is global process's model residual error of multivariable control system,To estimate in multivariable control system
Inverse, the G of the transfer function matrix of the interference model of metermod(z-1) be multivariable control system process model transmission function, y
It (t) is the output of process of multivariable control system, u (t) is that the process of multivariable control system inputs.
5. a kind of model mismatch diagnostic method of the multivariable control system of data-driven as claimed in claim 1 or 2, special
Sign is, when world model's quality index of the multivariable control system is less than 0.98, the whole mistake of multivariable control system
Journey model mismatch.
6. a kind of model mismatch diagnostic method of the multivariable control system of data-driven as claimed in claim 1 or 2, special
Sign is that when the overall model quality index of i-th of output channel of the multivariable control system is less than 0.98, i-th defeated
The process model mismatch in channel out.
7. a kind of model mismatch diagnostic method of the multivariable control system of data-driven as claimed in claim 1 or 2, special
Sign is, the partial model quality index of i-th of output channel are as follows:
Wherein, t is t-th of sampling instant,For the variance of the white noise of i-th of output channel of estimation,For the variance of the white noise of j-th of input channel of estimation,For the order of the linear regression model (LRM) of estimation,For the coefficient of the linear regression model (LRM) of estimation, ηL, ijFor i-th of output channel of j-th of multivariable control system input
Partial model quality index,
8. a kind of model mismatch diagnostic method of the multivariable control system of data-driven as claimed in claim 1 or 2, special
Sign is, when the partial model quality index of i-th of output channel is less than 0.98, the process mould of the input and output subchannel
Type mismatch.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810357053.7A CN108536127B (en) | 2018-04-20 | 2018-04-20 | A kind of model mismatch diagnostic method of the multivariable control system of data-driven |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810357053.7A CN108536127B (en) | 2018-04-20 | 2018-04-20 | A kind of model mismatch diagnostic method of the multivariable control system of data-driven |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108536127A CN108536127A (en) | 2018-09-14 |
CN108536127B true CN108536127B (en) | 2019-08-13 |
Family
ID=63479035
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810357053.7A Expired - Fee Related CN108536127B (en) | 2018-04-20 | 2018-04-20 | A kind of model mismatch diagnostic method of the multivariable control system of data-driven |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108536127B (en) |
Families Citing this family (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110090727B (en) * | 2019-04-26 | 2020-12-22 | 北矿智云科技(北京)有限公司 | Method, device, equipment and medium for processing operation data in ore grinding production |
CN110941797B (en) * | 2019-11-07 | 2023-04-07 | 中信银行股份有限公司 | Operation index monitoring and trend prediction system based on service index |
CN111176155B (en) * | 2019-12-20 | 2021-07-02 | 华中科技大学 | Process model mismatch detection method of closed-loop model predictive control system |
CN111679648B (en) * | 2020-05-22 | 2021-06-01 | 浙江浙能技术研究院有限公司 | Multivariate closed-loop control loop performance evaluation method based on Gaussian process regression |
CN113189967B (en) * | 2021-05-06 | 2022-05-27 | 郑州轻工业大学 | Control performance diagnosis method for semiconductor process batch control system |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102183699A (en) * | 2011-01-30 | 2011-09-14 | 浙江大学 | Method for model mismatching detection and positioning of multivariate predictive control system in chemical process |
CN104199296A (en) * | 2014-08-15 | 2014-12-10 | 上海交通大学 | Linear regression performance evaluation method with forgetting factor |
CN105700358A (en) * | 2016-03-14 | 2016-06-22 | 华中科技大学 | Modeling quality monitoring method for model predictive controller (MPC) with drift interference |
CN105807611A (en) * | 2016-03-05 | 2016-07-27 | 华中科技大学 | Detection method for mismatching of model of closed-loop control system and object |
CN107272640A (en) * | 2017-06-12 | 2017-10-20 | 华中科技大学 | A kind of modeling quality control method and system based on model predictive controller |
-
2018
- 2018-04-20 CN CN201810357053.7A patent/CN108536127B/en not_active Expired - Fee Related
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102183699A (en) * | 2011-01-30 | 2011-09-14 | 浙江大学 | Method for model mismatching detection and positioning of multivariate predictive control system in chemical process |
CN104199296A (en) * | 2014-08-15 | 2014-12-10 | 上海交通大学 | Linear regression performance evaluation method with forgetting factor |
CN105807611A (en) * | 2016-03-05 | 2016-07-27 | 华中科技大学 | Detection method for mismatching of model of closed-loop control system and object |
CN105700358A (en) * | 2016-03-14 | 2016-06-22 | 华中科技大学 | Modeling quality monitoring method for model predictive controller (MPC) with drift interference |
CN107272640A (en) * | 2017-06-12 | 2017-10-20 | 华中科技大学 | A kind of modeling quality control method and system based on model predictive controller |
Also Published As
Publication number | Publication date |
---|---|
CN108536127A (en) | 2018-09-14 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108536127B (en) | A kind of model mismatch diagnostic method of the multivariable control system of data-driven | |
CN108197380B (en) | Partial least square-based Gaussian process regression wastewater effluent index prediction method | |
US10919791B2 (en) | Intelligent identification method of sludge bulking based on type-2 fuzzy neural network | |
CN104699894B (en) | Gaussian process based on real-time learning returns multi-model Fusion Modeling Method | |
CN103927412B (en) | Instant learning debutanizing tower soft-measuring modeling method based on gauss hybrid models | |
CN108508865B (en) | A kind of fault detection method based on distributing OSC-PLS regression model | |
CN109407649B (en) | Fault type matching method based on fault characteristic variable selection | |
Tian et al. | A data-based approach for multivariate model predictive control performance monitoring | |
CN104914723A (en) | Industrial process soft measurement modeling method based on cooperative training partial least squares model | |
CN109492265B (en) | Wastewater effluent index prediction method based on dynamic nonlinear PLS soft measurement method | |
Zhao et al. | Soft sensor modeling of chemical process based on self-organizing recurrent interval type-2 fuzzy neural network | |
CN107403196B (en) | Method for predicting concentration of butane at bottom of debutanizer by instant learning modeling based on spectral clustering analysis | |
CN109409425B (en) | Fault type identification method based on neighbor component analysis | |
CN101556458B (en) | Automatic control algorithm for feeding vitriol in tap water by coagulation | |
CN110059824A (en) | A kind of neural net prediction method based on principal component analysis | |
CN108664009A (en) | Divided stages based on correlation analysis and fault detection method | |
CN110210687A (en) | A kind of Nonlinear Dynamic production process product quality prediction technique returned based on local weighted slow feature | |
CN105807611B (en) | A kind of model of closed-loop control system and the unmatched detection method of object | |
CN109145421B (en) | Time-space fuzzy modeling method applied to distributed parameter system | |
CN110032799A (en) | A kind of the angle similarity divided stages and monitoring method of microbiological pharmacy process | |
CN108345294B (en) | A kind of fault detection method based on distributing principal component regression model | |
CN105242572A (en) | Mixing identification method and system for thermal process of thermal power plant | |
CN111914887B (en) | Novel multi-mode chemical process abnormal state detection method | |
CN108646553A (en) | A method of statistics on-line monitoring closed-loop control system model quality | |
CN111429718A (en) | Intelligent detection system for road traffic safety |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20190813 Termination date: 20200420 |
|
CF01 | Termination of patent right due to non-payment of annual fee |