CN108153146A - A kind of polynary molten steel quality MFA control system and method for blast furnace - Google Patents
A kind of polynary molten steel quality MFA control system and method for blast furnace Download PDFInfo
- Publication number
- CN108153146A CN108153146A CN201711316435.7A CN201711316435A CN108153146A CN 108153146 A CN108153146 A CN 108153146A CN 201711316435 A CN201711316435 A CN 201711316435A CN 108153146 A CN108153146 A CN 108153146A
- Authority
- CN
- China
- Prior art keywords
- parameter
- molten steel
- model
- steel quality
- adaptive controller
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B13/00—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion
- G05B13/02—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric
- G05B13/0205—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric not using a model or a simulator of the controlled system
- G05B13/024—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric not using a model or a simulator of the controlled system in which a parameter or coefficient is automatically adjusted to optimise the performance
Abstract
The present invention provides a kind of polynary molten steel quality MFA control system and method for blast furnace.Set the polynary molten steel quality index desired value of blast furnace；The blast-furnace body production status parametric variable of several historical junctures and polynary molten steel quality indicator measurements of corresponding moment needed for model-free adaptive controller parameter correction are obtained, off-line correction is carried out to sensitive model-free adaptive controller parameter and pseudo- partial derivative initial value；Utilize the polynary molten steel quality index desired value of blast furnace of model-free adaptive controller On-line Control molten steel quality index tracking setting.The online recursion controller of the polynary molten steel quality of blast furnace is established with reference to MFA control technology；And the recursion subspace prediction model between the polynary molten steel quality of blast furnace and two controlled quentity controlled variables is established as parameter correction submodel, using multi-parameter sensitivity analysis, genetic algorithm off-line correction controller parameter and pseudo- partial derivative initial value, the current dynamic characteristic of blast furnace is adapted it to, realizes effective control to polynary molten steel quality.
Description
Technical field
The invention belongs to blast furnace process technical field of automatic control, more particularly to a kind of polynary molten steel quality of blast furnace is without mould
Type adaptive control system and method.
Background technology
The final purpose of blast furnace ironmaking mainly produces the good pig iron, and qualification is provided for downstream processes such as pneumatic steelmakings
Head product.Therefore, the control to blast furnace final products molten steel quality is generally also meant to be to the optimal control of blast furnace ironmaking.Mesh
Before, molten steel quality index generally use Si contents, molten iron temperature, S contents and P content are weighed.Wherein Si contents and molten iron temperature
Degree is the important parameter for characterizing blast furnace chemical heat and physical thermal respectively.Molten iron silicon content is always for reflecting in blast furnace for many years
The leading indicator of portion's Warm status, it adequately the chemistry in reacting furnace and physical state, molten iron temperature can stablize blast furnace suitable
Row and energy expenditure all have great influence.Therefore, strictly control molten iron silicon content and molten iron temperature are right in rational range
Blast furnace production process is of great significance.
Patent publication No. CN106681146A discloses " the blast furnace multiple-objection optimization based on BP neural network and genetic algorithm
Control algolithm ", it is defeated in order to control with blast energy, hot-blast pressure, hot blast temperature, cold flow, full tower pressure difference, oxygen enrichment percentage and coal ratio
Enter, using molten steel sulfur content, CO2 emissions and coke ratio as output variable, establish and train BP neural network, use NSGA-II
Multi-objective optimization algorithm finds out optimal control variable value.
Patent publication No. CN106249724A discloses " a kind of polynary molten steel quality forecast Control Algorithm of blast furnace and system ",
Using hot blast temperature, hot-blast pressure, oxygen enrichment percentage, injecting coal quantity is set as input, established by output of the polynary molten steel quality index of blast furnace
Least square method supporting vector machine regression model, and using this model as prediction model, to blast furnace, polynary molten steel quality index carries out pre-
Observing and controlling system solves optimum control amount with sequential quadratic programming algorithm.
Method in the method for above-mentioned patent report and other pertinent literatures is on the basis of the model established offline
The controller of foundation.However blast furnace process system is a uncertain dynamic time-varying system, the change of the grade of ore, operating mode
The variation of modeled shape parameter can be led to by changing, and therefore, off-line model can not reflect the dynamic of blast furnace different moments comprehensively, with
The controller of this Foundation is also just only applicable to specific blast furnace duty parameter, and the control that can not meet entire production process will
It asks, actual application value is relatively low.The data model computation complexity for the control method foundation that in addition, there will be is larger, calculates the time
It is longer, it is not suitable for On-line Control, it is difficult to applied to real industry spot.It is in conclusion special not yet both at home and abroad at present
The method for carrying out polynary dynamic On-line Control for blast furnace ironmaking process molten steel quality index S i contents and molten iron temperature.
Invention content
In view of the problems of the existing technology, the present invention provides a kind of polynary molten steel quality MFA control of blast furnace
System and method.
The technical scheme is that：
A kind of polynary molten steel quality MFA control method of blast furnace, including：The setting polynary molten steel quality of blast furnace refers to
Mark desired value；It further includes：
Obtain the blast-furnace body production status parameter of several historical junctures needed for model-free adaptive controller parameter correction
Variable and polynary molten steel quality indicator measurements of corresponding moment, to sensitive model-free adaptive controller parameter and pseudo- partial derivative
Initial value carries out off-line correction；
Utilize the polynary molten steel quality of blast furnace of model-free adaptive controller On-line Control molten steel quality index tracking setting
Index desired value：Model-free adaptive controller is using last moment output tracking error as input, with current time model-free
Adaptive controller output controlled quentity controlled variable increment is to set the increment of top injecting coal quantity and pressure difference as exporting, and is moved using output self feed back
The online recursion control molten steel quality index output of state.
It is described that off-line correction is carried out to sensitive model-free adaptive controller parameter and pseudo- partial derivative initial value, including：
Using the recursion Subspace algorithm based on forgetting factor, if using needed for model-free adaptive controller parameter correction
The dry blast-furnace body production status parametric variable of historical juncture and polynary molten steel quality indicator measurements of corresponding moment, establish parameter
Correct submodel；
Using at the beginning of each control parameter of multi-parameter sensitivity analysis technique analysis model-free adaptive controller and pseudo- partial derivative
It is worth sensitivity；
Genetic algorithm optimization correction of typist's errors is used to be analyzed to identify via multi-parameter sensitivity analysis technique sensitively to control
Parameter and pseudo- partial derivative initial value；
The polynary molten steel quality index tracking of model-free adaptive controller control parameter correction submodel output is set
The accumulative root-mean-square error at fixed output reference curve several moment carries out model-free adaptive controller as evaluation index
Evaluation：If the controller evaluation index value for adding up several moment is less than given threshold, current model-free adaptive controller
It meets the requirements；Otherwise current model-free adaptive controller is unsatisfactory for requiring, and reacquires model-free adaptive controller parameter
The blast-furnace body production status parametric variable of several historical junctures of blast furnace needed for correction and polynary molten steel quality index of corresponding moment
Measured value analyzes model-free adaptive controller parameter and pseudo- partial derivative initial value sensitivity, at the beginning of sensitive parameter and pseudo- partial derivative
Value carries out off-line correction.
It is described to establish parameter correction submodel, including：
The blast-furnace body production status parameter of historical junctures several needed for model-free adaptive controller parameter correction is become
Amount and corresponding moment polynary molten steel quality indicator measurements are divided into trained initial data and test initial data two parts；From instruction
Practice in initial data blast-furnace body production status parametric variable needed for extracting parameter correction submodel each moment training and corresponding
Moment polynary molten steel quality indicator measurements form multigroup training data, by each group training data by being arranged to make up training number constantly
According to collection；Extraction is for high needed for the test of each moment of test parameter correction submodel predictablity rate from test initial data
Furnace body production status parametric variable and polynary molten steel quality indicator measurements of corresponding moment form multigroup test data, by each group
Test data by being arranged to make up test data set constantly；
Based on training dataset, parameter correction submodel is built using subspace state space system identification, using add in forget because
The least square method of recursion of son solves parameter correction submodel coefficient matrix；
Evaluation index of the root-mean-square error as parameter correction submodel is introduced, based on test data set, if using working as
The root-mean-square error of each molten steel quality index that preceding parameter correction submodel estimates reaches given threshold, then the parameter school
Positive submodel is met the requirements；If the root-mean-square error of each molten steel quality index that parameter current correction submodel estimates
In, there is any one to be not up to given threshold, then rebuild parameter correction submodel；
Based on test data set, using the least square method of recursion for adding in forgetting factor, parameter correction submodel is solved
Coefficient matrix obtains final parameter correction submodel.
It is described that each control parameter of model-free adaptive controller and pseudo- local derviation are analyzed using multi-parameter sensitivity analysis technique
Number initial value sensitivity, including：
It constructs several groups and is uniformly distributed STOCHASTIC CONTROL parameter set, using parameter correction submodel as controlled device, with setting
Output reference curve target in order to control, using accumulative root mean square tracking error as target function value, STOCHASTIC CONTROL will be uniformly distributed
Each group control parameter assigns model-free adaptive controller respectively in parameter set, carries out several groups of Monte Carlo experiments；
Monte Carlo experimental results are sorted and are grouped, respectively drafting cumulative frequency curve, calculated curve separating degree,
Confirm sensitive control parameter and pseudo- partial derivative initial value.
The polynary molten iron of blast furnace using the tracking setting of model-free adaptive controller On-line Control molten steel quality index
Quality index desired value, including：
Current time blast-furnace body production status parametric variable is obtained in real time and obtains last moment polynary molten steel quality
Indicator measurements；
Establish the line of the relationship between the polynary molten steel quality index of description blast furnace and blast-furnace body production status parametric variable
Property equation；
Last moment polynary molten steel quality index output valve and the polynary molten steel quality index desired value of blast furnace are made the difference, it is and new
Collected online blast-furnace body production status parametric variable feeds back to model-free adaptive controller input terminal, and recursion update is pseudo-
The output of partial derivative matrix and model-free adaptive controller.
A kind of polynary molten steel quality MFA control system of blast furnace, including：Target setting unit：It is more to set blast furnace
First molten steel quality index desired value；It further includes：
Controller start unit：Obtain the height of blast furnace several historical junctures needed for model-free adaptive controller parameter correction
Furnace body production status parametric variable and polynary molten steel quality indicator measurements of corresponding moment, to sensitive model-free adaption control
Device parameter processed and pseudo- partial derivative initial value carry out off-line correction；
On-line Control unit：Utilize the blast furnace of model-free adaptive controller On-line Control molten steel quality index tracking setting
Polynary molten steel quality index desired value：Model-free adaptive controller is made with last moment polynary molten steel quality output tracking error
For input, controlled quentity controlled variable increment is exported as output using current time model-free adaptive controller, using output self feed back dynamic
Online recursion control molten steel quality index output.
The controller start unit, including：
Parameter correction submodel establishes module：Using the recursion Subspace algorithm based on forgetting factor, model-free is utilized
The blast-furnace body production status parametric variable of several historical junctures needed for adaptive controller parameter correction and corresponding moment are polynary
Molten steel quality indicator measurements establish parameter correction submodel；
Control parameter sensitivity analysis module：Model-free adaptive controller is analyzed using multi-parameter sensitivity analysis technique
Each control parameter and pseudo- partial derivative initial value sensitivity；
Sensitive parameter correction module：It is analyzed using genetic algorithm optimization correction of typist's errors via multi-parameter sensitivity analysis technique
It is confirmed as sensitive control parameter and pseudo- partial derivative initial value；
Control performance evaluation module：By the polynary iron of model-free adaptive controller control parameter correction submodel output
The accumulative root-mean-square error at output reference curve several moment of water quality figureofmerit tracking setting is as evaluation index, if accumulative
The controller evaluation index value at several moment is less than given threshold, then current model-free adaptive controller is met the requirements；Otherwise
Current model-free adaptive controller is unsatisfactory for requiring, and reacquires several needed for model-free adaptive controller parameter correction go through
The blast-furnace body production status parametric variable at history moment and polynary molten steel quality indicator measurements of corresponding moment, analysis model-free is certainly
Adaptive controller parameter and pseudo- partial derivative initial value sensitivity carry out off-line correction to sensitive parameter and pseudo- partial derivative initial value.
The parameter correction submodel establishes module, including：
Training and test data set constructing module：By the historical junctures several needed for model-free adaptive controller parameter correction
Blast-furnace body production status parametric variable and polynary molten steel quality indicator measurements of corresponding moment be divided into trained initial data
With test initial data two parts；The blast furnace sheet needed for each moment training of extracting parameter correction submodel from training initial data
Body production status parametric variable and polynary molten steel quality indicator measurements of corresponding moment form multigroup training data, and each group is trained
Data by being arranged to make up training dataset constantly；Extraction is for test parameter correction submodel prediction from test initial data
Blast-furnace body production status parametric variable and polynary molten steel quality index measurement of corresponding moment needed for the test of each moment of accuracy rate
Value forms multigroup test data, by each group test data by being arranged to make up test data set constantly；
Model training module：Based on training dataset, parameter correction submodel is built using subspace state space system identification, is adopted
With the least square method of recursion for adding in forgetting factor, parameter correction submodel coefficient matrix is solved；
Model evaluation module：Evaluation index of the root-mean-square error as parameter correction submodel is introduced, based on test number
According to collection, if the root-mean-square error of each molten steel quality index that parameter current correction submodel estimates reaches given threshold,
Then the parameter correction submodel is met the requirements；If each molten steel quality index that parameter current correction submodel estimates is equal
In square error, there is any one to be not up to given threshold, then rebuild parameter correction submodel；
Model construction module：Based on test data set, using the least square method of recursion for adding in forgetting factor, parameter is solved
Submodel coefficient matrix is corrected, obtains final parameter correction submodel.
The control parameter sensitivity analysis module, including：
Monte Carlo simulation modules：It constructs several groups and is uniformly distributed STOCHASTIC CONTROL parameter set, mould is assisted with parameter correction
Type is controlled device, reference curve target in order to control is exported with the polynary molten steel quality of setting, to add up root mean square tracking error
For target function value, each group control parameter in STOCHASTIC CONTROL parameter set will be uniformly distributed and assign MFA control respectively
Device carries out several groups of Monte Carlo experiments；
Sensitivity analysis module：Monte Carlo experimental results are sorted and are grouped, draw cumulative frequency curve respectively,
Calculated curve separating degree confirms sensitive control parameter and pseudo- partial derivative initial value.
The On-line Control unit, including：
Online data acquisition module：Current time blast-furnace body production status parametric variable and acquisition upper one are obtained in real time
Moment polynary molten steel quality indicator measurements；
Establishing equation module：It establishes between the polynary molten steel quality index of description blast furnace and blast-furnace body duty parameter variable
The linear equation of relationship；
Control module：By last moment polynary molten steel quality index output valve and the polynary molten steel quality index desired value of blast furnace
It makes the difference, with new collected online blast furnace operating mode parameter feedback to model-free adaptive controller input terminal, recursion update is pseudo- partially
The output of Jacobian matrix and model-free adaptive controller.
Advantageous effect：
The present invention makes blast-melted quality index quickly and accurately reach to provisioning request, there is stronger resistance external disturbance
Ability can improve molten steel quality in the actual production process, reduce energy consumption, save cost.According to blast furnace industry spot sensing
The practical polynary molten steel quality related data of blast furnace measured of device, it is more to establish a blast furnace with reference to MFA control technology
The online recursion controller of first molten steel quality (Si contents, molten iron temperature)；And establish height with reference to blast furnace recent history creation data
Recursion subspace prediction model between the polynary molten steel quality of stove and two controlled quentity controlled variables (pressure difference, setting injecting coal quantity), by this model
As parameter correction submodel, using multi-parameter sensitivity analysis technique, genetic algorithm optimization technology, established ginseng is utilized
Number correction submodel off-line correction controller parameter and pseudo- partial derivative initial value, adapt it to the current dynamic characteristic of blast furnace, realize
Effective control to polynary molten steel quality.The present invention has higher control accuracy, using model-free adaptive controller, tool
There is faster arithmetic speed, be suitable for real-time industrial process；Have the advantages that tracking performance is good, strong interference immunity.
Description of the drawings
Fig. 1 is the polynary molten steel quality MFA control system block diagram of blast furnace in the specific embodiment of the invention；
Fig. 2 is controller start unit block diagram in the specific embodiment of the invention；
Fig. 3 is that parameter correction submodel establishes module frame chart in the specific embodiment of the invention；
Fig. 4 is On-line Control unit block diagram in the specific embodiment of the invention；
Fig. 5 is the polynary molten steel quality MFA control method flow diagram of blast furnace in the specific embodiment of the invention；
Fig. 6 is step 2 flow chart in the specific embodiment of the invention；
Fig. 7 (a)~Fig. 7 (k) is λ in the specific embodiment of the invention, μ, η, ρ, a, b respectively1, b2, φ11, φ12, φ21,
φ22Cumulative frequency curve；
Fig. 8 (a), Fig. 8 (b) are the control effect of molten iron silicon content and molten iron temperature in the specific embodiment of the invention respectively
Figure, Fig. 8 (c), Fig. 8 (d) are the change curve for setting injecting coal quantity and pressure difference during control respectively, and Fig. 8 (e)~Fig. 8 (h) divides
It is not control pseudo- partial derivative change curve in the process；
Fig. 9 (a), Fig. 9 (b) are the control effect for molten iron silicon content and molten iron temperature in the specific embodiment of the invention respectively
Fruit is schemed, and Fig. 9 (c), Fig. 9 (d) are respectively pressure difference and set change curve of the injecting coal quantity during control, Fig. 9 (e)~Fig. 9 (h)
It is control pseudo- partial derivative change curve in the process respectively.
Specific embodiment
It elaborates below in conjunction with the accompanying drawings to the specific embodiment of the present invention.
Present embodiment provides a kind of polynary molten steel quality MFA control system of blast furnace, as shown in Figure 1, packet
It includes：
Target setting unit：Set the polynary molten steel quality index desired value of blast furnace；
It further includes：
Controller start unit：Obtain the height of blast furnace several historical junctures needed for model-free adaptive controller parameter correction
Furnace body production status parametric variable and polynary molten steel quality indicator measurements of corresponding moment, to sensitive model-free adaption control
Device parameter processed and pseudo- partial derivative initial value carry out off-line correction；
On-line Control unit：Utilize the blast furnace of model-free adaptive controller On-line Control molten steel quality index tracking setting
Polynary molten steel quality index desired value：Model-free adaptive controller is made with last moment polynary molten steel quality output tracking error
For input, controlled quentity controlled variable increment is exported as output using current time model-free adaptive controller, using output self feed back dynamic
Online recursion control molten steel quality index output.
The controller start unit, as shown in Fig. 2, including：
Parameter correction submodel establishes module：Using the recursion Subspace algorithm based on forgetting factor, model-free is utilized
The blast-furnace body production status parametric variable of several historical junctures needed for adaptive controller parameter correction and corresponding moment are polynary
Molten steel quality indicator measurements establish parameter correction submodel；
Control parameter sensitivity analysis module：Model-free adaptive controller is analyzed using multi-parameter sensitivity analysis technique
Each control parameter and pseudo- partial derivative initial value sensitivity；
Sensitive parameter correction module：It is analyzed using genetic algorithm optimization correction of typist's errors via multi-parameter sensitivity analysis technique
It is confirmed as sensitive control parameter and pseudo- partial derivative initial value；
Control performance evaluation module：By the polynary iron of model-free adaptive controller control parameter correction submodel output
The accumulative root-mean-square error at output reference curve several moment of water quality figureofmerit tracking setting is as evaluation index, to model-free
Adaptive controller is evaluated：If the controller evaluation index value for adding up several moment is less than given threshold, currently without
Model self-adapted control device is met the requirements；Otherwise current model-free adaptive controller is unsatisfactory for requiring, and reacquires model-free
The blast-furnace body production status parametric variable of several historical junctures needed for adaptive controller parameter correction and corresponding moment are polynary
Molten steel quality indicator measurements analyze model-free adaptive controller parameter and pseudo- partial derivative initial value sensitivity, to sensitive parameter
And pseudo- partial derivative initial value carries out off-line correction.
The parameter correction submodel establishes module, as shown in figure 3, including：
Training and test data set constructing module：By the historical junctures several needed for model-free adaptive controller parameter correction
Blast-furnace body production status parametric variable and polynary molten steel quality indicator measurements of corresponding moment be divided into trained initial data
With test initial data two parts；The blast furnace sheet needed for each moment training of extracting parameter correction submodel from training initial data
Body production status parametric variable and polynary molten steel quality indicator measurements of corresponding moment form multigroup training data, and each group is trained
Data by being arranged to make up training dataset constantly；Extraction is for test parameter correction submodel prediction from test initial data
Blast-furnace body production status parametric variable and polynary molten steel quality index measurement of corresponding moment needed for the test of each moment of accuracy rate
Value forms multigroup test data, by each group test data by being arranged to make up test data set constantly；
Model training module：Based on training dataset, parameter correction submodel is built using subspace state space system identification, is adopted
With the least square method of recursion for adding in forgetting factor, parameter correction submodel coefficient matrix is solved；
Model evaluation module：Evaluation index of the root-mean-square error as parameter correction submodel is introduced, based on test number
According to collection, if the root-mean-square error of each molten steel quality index that parameter current correction submodel estimates reaches given threshold,
Then the parameter correction submodel is met the requirements；If each molten steel quality index that parameter current correction submodel estimates is equal
In square error, there is any one to be not up to given threshold, then rebuild parameter correction submodel；
Model construction module：Based on test data set, using the least square method of recursion for adding in forgetting factor, parameter is solved
Submodel coefficient matrix is corrected, obtains final parameter correction submodel.
The control parameter sensitivity analysis module, including：
Monte Carlo simulation modules：It constructs several groups and is uniformly distributed STOCHASTIC CONTROL parameter set, mould is assisted with parameter correction
Type is controlled device, reference curve target in order to control is exported with the polynary molten steel quality of setting, to add up root mean square tracking error
For target function value, each group control parameter in STOCHASTIC CONTROL parameter set will be uniformly distributed and assign MFA control respectively
Device carries out several groups of Monte Carlo experiments；
Sensitivity analysis module：Monte Carlo experimental results are sorted and are grouped, draw cumulative frequency curve respectively,
Calculated curve separating degree confirms sensitive control parameter and pseudo- partial derivative initial value.
The On-line Control unit, as shown in figure 4, including：
Online data acquisition module：Current time blast-furnace body production status parametric variable and acquisition upper one are obtained in real time
Moment polynary molten steel quality indicator measurements；
Establishing equation module：Establish the description polynary molten steel quality index of blast furnace (molten iron silicon content, molten iron temperature) and blast furnace
The linear equation of relationship between ontology duty parameter variable (setting injecting coal quantity, pressure difference)；
Control module：By last moment polynary molten steel quality index output valve and the polynary molten steel quality index desired value of blast furnace
It makes the difference, model-free adaptive controller input terminal is fed back to new collected online blast-furnace body production status parametric variable,
The output of the pseudo- partial derivative matrix of recursion update and model-free adaptive controller.
A kind of polynary molten steel quality MFA control method of blast furnace, as shown in figure 5, including：
Step 1, setting the polynary molten steel quality index desired value of blast furnace, the polynary molten steel quality index include silicone content,
Molten iron temperature.
Step 2, the blast-furnace body production work for obtaining several historical junctures needed for model-free adaptive controller parameter correction
Condition parametric variable and polynary molten steel quality indicator measurements of corresponding moment, to sensitive model-free adaptive controller parameter and puppet
Partial derivative initial value carries out off-line correction.
Flow as shown in Figure 6 specifically includes：
Step 2-1, the blast-furnace body production of several historical junctures needed for model-free adaptive controller parameter correction is obtained
Duty parameter variable and polynary molten steel quality indicator measurements of corresponding moment, line number of going forward side by side Data preprocess；
The blast-furnace body at recent 450 moment needed for model-free adaptive controller parameter correction is obtained in present embodiment
Production status parametric variable and polynary molten steel quality indicator measurements of corresponding moment；Blast-furnace body production status needed for controller is joined
Number variable, including setting injecting coal quantity and pressure difference；The polynary molten steel quality indicator measurements of blast furnace, including silicone content, molten iron temperature；In advance
Processing includes：Using noise spike filtering algorithm cancelling noise spike saltus step data；Number is eliminated using gliding smoothing filtering algorithm
High frequency measurement noise in；Data are normalized；
Step 2-2, using the recursion Subspace algorithm based on forgetting factor, model-free adaptive controller parameter school is utilized
The blast-furnace body production status parametric variable of just required several historical junctures and polynary molten steel quality indicator measurements of corresponding moment,
Establish parameter correction submodel；
Step 2-2-1, the blast-furnace body of historical junctures several needed for model-free adaptive controller parameter correction is produced
Duty parameter variable and polynary molten steel quality indicator measurements of corresponding moment are divided into trained initial data and test initial data
Two parts；The blast-furnace body production status parameter needed for each moment training of extracting parameter correction submodel from training initial data
Variable and polynary molten steel quality indicator measurements of corresponding moment form multigroup training data, and each group training data is arranged by the moment
Composing training data set；Extraction corrects each moment of submodel predictablity rate for test parameter from test initial data
Blast-furnace body production status parametric variable needed for test and polynary molten steel quality indicator measurements of corresponding moment form multigroup test
Data, by each group test data by being arranged to make up test data set constantly.
200 groups of data are used as test as training initial data and 100 groups of data before being extracted from pretreated data
Initial data two parts；The blast-furnace body production status parametric variable includes setting injecting coal quantity, pressure difference；Polynary molten steel quality refers to
Mapping magnitude includes silicone content measured value, molten iron temperature measured value；
6 input variables needed for training parameter correction submodel are as follows：
Current time setting injecting coal quantity u1(t), m3/h；
Current time pressure difference u2(t), KPa；
Last moment setting injecting coal quantity u1(t-1), m3/h；
Last moment pressure difference u2(t-1), KPa；
Last moment silicone content measured value y1(t-1), %；
Last moment molten iron temperature measured value y2(t-1), %.
2 output variables needed for training parameter correction submodel are as follows：
Current time silicone content detected value y1(t), %；；
Current time molten iron temperature detected value y2DEG C (t),.
Then t groups training data is：
X (t)=[y1(t-1), y2(t-1), u1(t-1), u2(t-1), u1(t), u2(t)]T。
Step 2-2-2, based on training dataset, parameter correction submodel is built using subspace state space system identification, is used
The least square method of recursion of forgetting factor is added in, solves parameter correction submodel coefficient matrix；
Using subspace state space system identification build parameter correction submodel beIt is wherein auxiliary for parameter correction
Model coefficient matrix is helped, since blast furnace ironmaking process is the slow time-varying process of an operating mode gradual change, parameter correction assists mould
The solution of type coefficient matrix L in controller parameter optimization process, is utilized using the least square method of recursion for adding in forgetting factor
Blast-furnace body production status parametric variable iteration updates model parameter, is updated by least square method of recursion iteration, L ∈ Rl ×(2*m+l), m, l are respectively that model-free adaptive controller outputs and inputs dimension(t) refer to for the polynary molten steel quality of t moment
Mark estimated value；Model-free adaptive controller after optimization adapts to the slow time-varying characteristic of blast furnace ironmaking process, is as follows：
Step 2-2-2-1, enable recursion iterations N=1, forgetting factor λ ∈ (0,1], set coefficient matrix initial valueCovariance matrix initial value
Step 2-2-2-2, it definesThen gain matrix Kt+1It is as follows：
Step 2-2-2-3, coefficient matrix Lt+1It is updated by following formula：
Step 2-2-2-4, covariance matrix Pt+1It is updated by following formula：
Step 2-2-2-5, t=t+1 is enabled, step 2-2-2-2 is jumped to, repeats above step, until recursion iteration
Number exceeds data amount check in data set, jumps out cycle, obtains coefficient matrix L=LN。
Step 2-2-3, evaluation index of the root-mean-square error as parameter correction submodel is introduced, based on test data
Collection, if reaching setting threshold using the root-mean-square error of each molten steel quality index that parameter current correction submodel estimates
Value, then the parameter correction submodel is met the requirements；If each molten steel quality index that parameter current correction submodel estimates
Root-mean-square error in, there is any one to be not up to given threshold, then rebuild parameter correction submodel；
Root-mean-square error is defined as follows formula：
Step 2-2-4, based on test data set, using the least square method of recursion for adding in forgetting factor, parameter school is solved
Positive submodel coefficient matrix, obtains final parameter correction submodel：In the parameter school solved based on training dataset
Positive submodel coefficient matrix LNOn the basis of recursion go out parameter correction submodel coefficient matrix [LN+1, LN+2..., LN+250], N
=200, this parameter set is final parameter correction submodel；
Step 2-3, using each control parameter of multi-parameter sensitivity analysis technique analysis model-free adaptive controller and puppet
Partial derivative initial value sensitivity.
Model-free adaptive controller, which contains more variable control parameters, to be needed to correct, and is introduced into MFA control
Multiple pseudo- partial derivative initial values can be increased in device again.Through experimental examination, for model-free adaptive controller, and not all model-free
Adaptive controller parameter and pseudo- partial derivative initial value have larger impact to model-free adaptive controller performance, to save the time
Cost makes offline parameter correction time shorten as possible, and model-free adaptive controller puppet partial derivative is anti-during so as to make On-line Control
It is adaptive to solve model-free closer to the current dynamic working point of blast furnace using multi-parameter sensitivity analysis technique for the dynamic characteristic reflected
The problem of answering parameter correction in controller analyzes model-free adaptive controller parameter and pseudo- partial derivative initial value sensitivity, will be clever
The low parameter of sensitivity and pseudo- partial derivative initial value are rejected, so as to improve the control accuracy of model-free adaptive controller and efficiency.
Need the control parameter for carrying out the model-free adaptive controller of sensitivity analysis as follows：
For punishing weight factor λ, the λ ＞ 0 of the model-free adaptive controller output excessive variation of controlled quentity controlled variable calculated value；
For punishing weight factor μ, the μ ＞ 0 of the excessive variation of PJM estimated values；
Model-free adaptive controller output controlled quentity controlled variable update step factor ρ, ρ ∈ (0,1]；
Pseudo- partial derivative update step factor η, η ∈ (0,2]；
Pseudo- partial derivative value limits parameter alpha, b1, b2, α ＞ 1, b2＞ b1(2α+1)；
Pseudo- partial derivative initial value φ11, φ12, φ21, φ22。
The step 2-3 includes：
Step 2-3-1, several groups are constructed and is uniformly distributed STOCHASTIC CONTROL parameter set, using parameter correction submodel as controlled pair
As with the output reference curve of setting target in order to control, using accumulative root mean square tracking error as target function value, will be uniformly distributed
Each group control parameter assigns model-free adaptive controller respectively in STOCHASTIC CONTROL parameter set, carries out several groups of Monte Carlo
Experiment；
The step 2-3-1 includes：
Step 2-3-1-1, the value range of each control parameter is set；
Control parameter value range is as follows：
| Parameter | Value lower limit | The value upper limit | Parameter | Value lower limit | The value upper limit |
| λ | 0 | 20 | b2 | 0 | 1000 |
| μ | 0 | 20 | φ11 | -20 | 20 |
| ρ | 0 | 1 | φ12 | —20 | 20 |
| η | 0 | 2 | φ21 | -20 | 20 |
| α | 1 | 20 | φ22 | -20 | 20 |
| b1 | 0 | 20 |
B in upper table1, b2, the value of α also meets following inequality relation：
b2＞ b1(2α+1)(m-1) (5)
Step 2-3-1-2, it is each control parameter in its value range, generates 500 and obey equally distributed independences
Random number forms 500 groups of STOCHASTIC CONTROL parameter sets；
Step 2-3-1-3, using 500 groups of STOCHASTIC CONTROL parameter sets of generation, it is separately operable model-free adaptive controller
At blast furnace polynary molten steel quality index output 200 moment of desired value of control parameter correction submodel tracking setting, calculate mesh
Offer of tender numerical value；
Object function is：
Wherein eiTracking error vector for the i moment.
Step 2-3-2, Monte Carlo experimental results are sorted and are grouped, draw cumulative frequency curve respectively, calculated bent
Line separating degree confirms sensitive control parameter and pseudo- partial derivative initial value.
The step 2-3-2 includes：
Step 2-3-2-1, according to target function value, 500 groups of STOCHASTIC CONTROL parameter sets are divided into two groups, it is respectively accumulative with
Track square error " permission " control parameter collection and accumulative tracking square error " do not allow " control parameter collection.
500 groups of STOCHASTIC CONTROL parameter set groupings are according to may be used " subjective polynary molten steel quality error criterion ", and " subjectivity is more
First molten steel quality error criterion " is the median of the target function value of 500, if target function value is more than " subjective polynary iron
Water quality error criterion ", then the control parameter group is divided into " not allowing " control parameter collection；, whereas if object function
Value is less than " subjective polynary molten steel quality error criterion ", then corresponding control parameter group is divided into " permission " control parameter
Collection.
Step 2-3-2-2, to each control parameter, " permission " control parameter collection and " not allowing " control parameter are calculated respectively
The cumulative frequency of centralized Control parameter values, draws cumulative frequency curve, and the present embodiment paints cumulative frequency distribution curve such as Fig. 7
(a) shown in~(k)；The distribution situation of two curves is observed, calculates the separating degree of two curves, point of two cumulative frequency curves
From the level of sensitivity that degree represents parameter.
The calculation formula of separating degree is as follows：
Wherein,And yiIt is the cumulative frequency number that control parameter y is divided into " permission " and " not allowing " control parameter collection respectively
Value,It is the average value for the cumulative frequency numerical value that control parameter y is divided into " permission " control parameter collection.
Step 2-3-2-3, sensitive control parameter and pseudo- partial derivative initial value are determined：Choose the tired of " permission " control parameter collection
Meter frequency curve is detached with the cumulative frequency curve of " not allowing " control parameter collection and the control parameter of separating degree DS ＜ 0.9 is spirit
Quick control parameter, corresponding puppet partial derivative initial value are sensitive pseudo- partial derivative initial value；" permission " control parameter collection adds up
The cumulative frequency curve of frequency curve and " not allowing " control parameter collection does not detach or the control parameter of separating degree DS >=0.9 is not
Sensitive control parameter, corresponding puppet partial derivative initial value are insensitive pseudo- partial derivative initial value.
Step 2-4, genetic algorithm optimization correction of typist's errors is used to be analyzed to identify via multi-parameter sensitivity analysis technique as spirit
Quick control parameter and pseudo- partial derivative initial value.
Model-free adaptive controller is introduced into the polynary molten steel quality norm controlling of blast furnace per se with more control parameter
Find there is larger impact to the control performance of model-free adaptive controller afterwards in the process, due to lacking engineering experience,
The parameter value of his process industry system again and does not apply to, therefore can not directly apply to blast-melted quality index control, needs
Correct model-free adaptive controller parameter.Confirmed using genetic algorithm optimization correction of typist's errors via multi-parameter sensitivity analysis
For sensitive parameter and pseudo- partial derivative initial value.
The step 2-4 includes：
Step 2-4-1, with reference to the polynary molten steel quality norm controlling target of blast furnace, evaluation mark is provided for genetic algorithm optimizing
Standard constructs fitness function.
Fitness function is configured to the accumulative square error of polynary molten steel quality Indicator setpoint tracking output, following institute
Show：
Wherein, y* is the polynary molten steel quality index desired value at current time.
If it enables
Param=[λ, μ, ρ, η, α, b1, b2, φ11, φ12, φ21, φ22]T
LB=[0,0,0,0,1,0,0, -20, -20, -20, -20]T (9)
UB=[20,20,1,2,20,20,1000,20,20,20,20]T
The wherein lower limit vector sum upper limit vector of LB, UB parameter value range in order to control, numerical value is derived from step 2-3-1-1
Control parameter value range table.
It is to solve following optimization problem to find model-free adaptive controller optimized parameter：
Step 2-4-2, by the way that the ga () function in Matlab tool boxes is called to be asked come the optimization in solution procedure 2-4-1
Topic, call format are：
[x, fval]=ga (fitnessfcn, nvars, A, b, Aeq, beq, LB, UB, nonlcon)
The reference page being specifically defined referring to ga in Matlab of corresponding entry.
By calling above-mentioned ga () function, control parameter and puppet that model-free adaptive controller performance is optimal are acquired
Control parameter after correction and pseudo- partial derivative initial value are passed to model-free adaptive controller by the estimated value of partial derivative initial value,
For the polynary molten steel quality index output of model-free adaptive controller On-line Control.
Step 2-5, the model-free adaptive controller corresponding to parameter current correction submodel will will be utilized to estimate
Each molten steel quality index root-mean-square error as evaluation index, model-free adaptive controller is evaluated：It is if tired
The controller evaluation index value for counting several moment is less than given threshold, then current model-free adaptive controller is met the requirements；It is no
Then current model-free adaptive controller is unsatisfactory for requiring, and reacquires blast furnace needed for model-free adaptive controller parameter correction
The blast-furnace body production status parametric variable of several historical junctures and polynary molten steel quality indicator measurements of corresponding moment analyze nothing
Model self-adapted control device parameter and pseudo- partial derivative initial value sensitivity carry out offline school to sensitive parameter and pseudo- partial derivative initial value
Just.
Step 3, the polynary iron of blast furnace using the tracking setting of model-free adaptive controller On-line Control molten steel quality index
Water quality figureofmerit desired value：Model-free adaptive controller is using last moment output tracking error as input, with current time
Model-free adaptive controller output controlled quentity controlled variable increment is to set the increment of injecting coal quantity and pressure difference as output, reflexive using exporting
The online recursion control molten steel quality index output of feedback dynamic.
The step 3 includes：
Step 3-1, on obtaining current time blast-furnace body production status parametric variable in real time using sensor and obtaining
One moment polynary molten steel quality indicator measurements, line number of going forward side by side Data preprocess；
Pretreatment includes：Using noise spike filtering algorithm cancelling noise spike saltus step data；It is filtered using gliding smoothing
Algorithm eliminates the high frequency measurement noise in data；Data are normalized；
Step 3-2, in the blast furnace system dynamic working point of t moment, the description polynary molten steel quality index of blast furnace and height are established
The linear equation of relationship between furnace body production status parametric variable：
Δ y (t)=Φ (t) × Δ u (t) (11)
Wherein,For pseudo- partial derivative matrix；Δ y (t)=[Δ y1(t)Δy2(t)]TIt is exported for t moment
The polynary molten steel quality index increment of blast furnace, Δ yi(t)=yi(t)-yi(t-1), i=1,2, y1For molten iron silicon content, y2For iron
Coolant-temperature gage；Δ u (t)=[Δ u1(t)Δu2(t)]TIncrement, Δ u are measured in order to controli(t)=ui(t)-ui(t-1), i=1,2, u1For
Set injecting coal quantity, u2For pressure difference.
Step 3-3, by last moment polynary molten steel quality index output valve and the polynary molten steel quality index desired value of blast furnace
It makes the difference, that is, seeks e (t)=y* (t+1)-y (t), wherein, y* (t)=[y1* (t), y2*(t)]TThe polynary molten iron of blast furnace for t moment
Quality index desired value, y1* (t) be molten iron silicon content desired value, y2* (t) is molten iron temperature desired value.By error e (t) and newly
Collected online blast-furnace body production status parametric variable feeds back to model-free adaptive controller input terminal, and recursion update is pseudo-
The output of partial derivative matrix and model-free adaptive controller.
The step 3-3 is specifically performed as follows：
Step 3-3-1, increased using last moment puppet partial derivative numerical value, last moment polynary molten steel quality index output valve
Amount, last moment controlled quentity controlled variable increment carry out the pseudo- partial derivative matrix of recursion update；The molten iron temperature predicted with model-free adaptive controller
Square error between degree, molten iron silicon content output increment increment and actual measurement molten iron temperature, molten iron silicon content output increment is damage
Function is lost, minimizes loss function so that pseudo- partial derivative matrix can more accurately reflect current time blast furnace system dynamic duty
The equivalent linearity characteristic of point, so as to which the update for measuring (setting injecting coal quantity, pressure difference) in order to control provides more accurately object information；In order to
Make algorithm that can more improve the dynamic characteristic of ground adaptive system Different Dynamic operating point with more generality, introduce pseudo- partial derivative
Square increment as penalty term；
Therefore, the criterion function of pseudo- Partial derivative estimation is as follows：
Wherein,Pseudo- Partial derivative estimation value for the k moment.
To criterion function derivation and it is enabled to be equal to 0 to solve minJ (Φ (t)), it is public finally to acquire pseudo- partial derivative iteration update
Formula is as follows：
Step 3-3-2, the pseudo- partial derivative of algorithm reset mechanism resetting is introduced, with the pseudo- partial derivative tracking blast furnace system time-varying of enhancing
The ability of parameter；
The step 3-3-2 includes：
Step 3-3-2-1, pseudo- partial derivative matrix the elements in a main diagonal is checkedWhether following item is met
Part：
(1) more than absolute minimum circle, i.e.,
(2) beyond relative maximum boundary, i.e.,
(3) it is different with the pseudo- partial derivative initial value symbol after correction, i.e.,Wherein,For
Pseudo- partial derivative initial value after correction, sign () are sign function.
When meeting any one in above three, resetIt enables
Step 3-3-2-2, the non-the elements in a main diagonal of pseudo- partial derivative matrix is checked
Whether following condition is met：
(1) more than absolute minimum circle, i.e.,
(2) it is different with the pseudo- partial derivative initial value symbol after correction, i.e.,
When meeting any one in above two, resetIt enables
Step 3-3-3, last moment model-free adaptive controller output controlled quentity controlled variable, updated pseudo- partial derivative square are utilized
Battle array, last moment polynary molten steel quality index output valve, the polynary molten steel quality index desired value of current time blast furnace carry out recursion more
Newly my model self-adapted control device exports controlled quentity controlled variable；Using the square error of polynary molten steel quality index output valve and desired value as
Loss function minimizes loss function so that silicone content and molten iron temperature reach desired value；In order to avoid model-free adaption control
Device output controlled quentity controlled variable variation processed is excessive, and real blast furnace system executing agency can not complete, and it is defeated to introduce model-free adaptive controller
Square increment gone out is as penalty term；
Therefore, the criterion function of model-free adaptive controller output controlled quentity controlled variable estimation is as follows：
J (u (t))=| | e (t) | |2+λ||u(t)-u(t-1)||2 (14)
To criterion function derivation and it is enabled to be equal to 0 to solve minJ (u (t)), finally acquire the online recursion update of controlled quentity controlled variable
Formula is as follows：
Step 3-4, model-free adaptive controller output controlled quentity controlled variable is applied in executing agency, into subsequent time,
Return to step 3-1.
In order to verify the performance of the polynary molten steel quality MFA control method and system of blast furnace in the present invention, carry out
Setting value tracking and interference--free experiments, Fig. 8 (a), Fig. 8 (b) be respectively in the specific embodiment of the invention molten iron silicon content and
The control effect figure of molten iron temperature, Fig. 8 (c), Fig. 8 (d) are the variation song for setting injecting coal quantity and pressure difference during control respectively
Line, Fig. 8 (e)~Fig. 8 (h) are pseudo- partial derivative change curve during control respectively；Fig. 9 (a), Fig. 9 (b) are tool of the present invention respectively
It is molten iron silicon content and the control effect figure of molten iron temperature in body embodiment, Fig. 9 (c), Fig. 9 (d) are pressure difference and setting respectively
Change curve of the injecting coal quantity during control, Fig. 9 (e)~Fig. 9 (h) are pseudo- partial derivative change curve during control respectively.
It can be seen that the polynary molten steel quality index MFA control method of blast furnace in the present invention is with good
Setting value tracking performance and interference free performance.
It is understood that above with respect to the specific descriptions of the present invention, it is merely to illustrate the present invention and is not limited to this
The described technical solution of inventive embodiments, it will be understood by those of ordinary skill in the art that, still the present invention can be carried out
Modification or equivalent replacement, to reach identical technique effect；As long as meeting using needs, all protection scope of the present invention it
It is interior.
Claims (10)
1. a kind of polynary molten steel quality MFA control method of blast furnace, including：Set the polynary molten steel quality index of blast furnace
Desired value；
It is characterized in that, it further includes：
Obtain the blast-furnace body production status parametric variable of several historical junctures needed for model-free adaptive controller parameter correction
And polynary molten steel quality indicator measurements of corresponding moment, to sensitive model-free adaptive controller parameter and pseudo- partial derivative initial value
Carry out off-line correction；
Utilize the polynary molten steel quality index of blast furnace of model-free adaptive controller On-line Control molten steel quality index tracking setting
Desired value：Model-free adaptive controller is adaptive with current time model-free using last moment output tracking error as input
Controller is answered to export, and controlled quentity controlled variable increment sets injecting coal quantity and the increment of pressure difference is used as output, online using output self feed back dynamic
Recursion control molten steel quality index output.
2. according to the method described in claim 1, it is characterized in that, it is described to sensitive model-free adaptive controller parameter and
Pseudo- partial derivative initial value carries out off-line correction, including：
Using the recursion Subspace algorithm based on forgetting factor, gone through using several needed for model-free adaptive controller parameter correction
The blast-furnace body production status parametric variable at history moment and polynary molten steel quality indicator measurements of corresponding moment, establish parameter correction
Submodel；
Using each control parameter of multi-parameter sensitivity analysis technique analysis model-free adaptive controller and pseudo- partial derivative initial value spirit
Sensitivity；
Genetic algorithm optimization correction of typist's errors is used to be analyzed to identify via multi-parameter sensitivity analysis technique as sensitive control parameter
And pseudo- partial derivative initial value；
By the polynary molten steel quality index tracking setting of model-free adaptive controller control parameter correction submodel output
The accumulative root-mean-square error at reference curve several moment is exported as evaluation index, model-free adaptive controller is commented
Valency：If the controller evaluation index value for adding up several moment is less than given threshold, current model-free adaptive controller is expired
Foot requirement；Otherwise current model-free adaptive controller is unsatisfactory for requiring, and reacquires model-free adaptive controller parameter school
The blast-furnace body production status parametric variable of just required blast furnace several historical junctures and polynary molten steel quality index of corresponding moment are surveyed
Magnitude analyzes model-free adaptive controller parameter and pseudo- partial derivative initial value sensitivity, to sensitive parameter and pseudo- partial derivative initial value
Carry out off-line correction.
3. according to the method described in claim 2, it is characterized in that, described establish parameter correction submodel, including：
By the blast-furnace body production status parametric variable of historical junctures several needed for model-free adaptive controller parameter correction and
Corresponding moment polynary molten steel quality indicator measurements are divided into trained initial data and test initial data two parts；It is former from training
Blast-furnace body production status parametric variable and corresponding moment needed for each moment training of beginning extracting data parameter correction submodel
Polynary molten steel quality indicator measurements form multigroup training data, by each group training data by being arranged to make up training data constantly
Collection；Extraction is for blast furnace needed for the test of each moment of test parameter correction submodel predictablity rate from test initial data
Ontology production status parametric variable and polynary molten steel quality indicator measurements of corresponding moment form multigroup test data, and each group is surveyed
Examination data by being arranged to make up test data set constantly；
Based on training dataset, parameter correction submodel is built using subspace state space system identification, using addition forgetting factor
Least square method of recursion solves parameter correction submodel coefficient matrix；
Evaluation index of the root-mean-square error as parameter correction submodel is introduced, based on test data set, if utilizing current ginseng
The root-mean-square error of each molten steel quality index that number correction submodel estimates reaches given threshold, then the parameter correction is auxiliary
Model is helped to meet the requirements；If in the root-mean-square error of each molten steel quality index that parameter current correction submodel estimates, have
Any one is not up to given threshold, then rebuilds parameter correction submodel；
Based on test data set, using the least square method of recursion for adding in forgetting factor, parameter correction submodel coefficient is solved
Matrix obtains final parameter correction submodel.
4. according to the method described in claim 2, it is characterized in that, described analyzed using multi-parameter sensitivity analysis technique without mould
Each control parameter of type adaptive controller and pseudo- partial derivative initial value sensitivity, including：
It constructs several groups and is uniformly distributed STOCHASTIC CONTROL parameter set, using parameter correction submodel as controlled device, with the defeated of setting
Go out reference curve target in order to control, using accumulative root mean square tracking error as target function value, STOCHASTIC CONTROL parameter will be uniformly distributed
Each group control parameter is concentrated to assign model-free adaptive controller respectively, carries out several groups of Monte Carlo experiments；
Monte Carlo experimental results are sorted and are grouped, draw cumulative frequency curve respectively, calculated curve separating degree confirms
Sensitive control parameter and pseudo- partial derivative initial value.
5. according to the method described in claim 1, it is characterized in that, described utilize model-free adaptive controller On-line Control iron
The polynary molten steel quality index desired value of blast furnace of water quality figureofmerit tracking setting, including：
Current time blast-furnace body production status parametric variable is obtained in real time and obtains last moment polynary molten steel quality index
Measured value；
Establish the linear side of the relationship between the polynary molten steel quality index of description blast furnace and blast-furnace body production status parametric variable
Journey；
Last moment polynary molten steel quality index output valve and the polynary molten steel quality index desired value of blast furnace are made the difference, with new acquisition
To online blast-furnace body production status parametric variable feed back to model-free adaptive controller input terminal, the pseudo- local derviation of recursion update
The output of matrix number and model-free adaptive controller.
6. a kind of polynary molten steel quality MFA control system of blast furnace, including：
Target setting unit：Set the polynary molten steel quality index desired value of blast furnace；
It is characterized in that, it further includes：
Controller start unit：Obtain the blast furnace sheet of blast furnace several historical junctures needed for model-free adaptive controller parameter correction
Body production status parametric variable and polynary molten steel quality indicator measurements of corresponding moment, to sensitive model-free adaptive controller
Parameter and pseudo- partial derivative initial value carry out off-line correction；
On-line Control unit：Blast furnace using the tracking setting of model-free adaptive controller On-line Control molten steel quality index is polynary
Molten steel quality index desired value：Model-free adaptive controller is using last moment polynary molten steel quality output tracking error as defeated
Enter, controlled quentity controlled variable increment is exported as output using current time model-free adaptive controller, it is online using output self feed back dynamic
Recursion control molten steel quality index output.
7. system according to claim 6, which is characterized in that the controller start unit, including：
Parameter correction submodel establishes module：It is adaptive using model-free using the recursion Subspace algorithm based on forgetting factor
Answer the blast-furnace body production status parametric variable of several historical junctures and polynary molten iron of corresponding moment needed for controller parameter correction
Quality index measured value establishes parameter correction submodel；
Control parameter sensitivity analysis module：It is respectively controlled using multi-parameter sensitivity analysis technique analysis model-free adaptive controller
Parameter processed and pseudo- partial derivative initial value sensitivity；
Sensitive parameter correction module：It is analyzed to identify using genetic algorithm optimization correction of typist's errors via multi-parameter sensitivity analysis technique
For sensitive control parameter and pseudo- partial derivative initial value；
Control performance evaluation module：By the polynary molten iron matter of model-free adaptive controller control parameter correction submodel output
The accumulative root-mean-square error at output reference curve several moment of figureofmerit tracking setting is adaptive to model-free as evaluation index
Controller is answered to be evaluated：If the controller evaluation index value for adding up several moment is less than given threshold, current model-free
Adaptive controller is met the requirements；Otherwise current model-free adaptive controller is unsatisfactory for requiring, and it is adaptive to reacquire model-free
Answer the blast-furnace body production status parametric variable of blast furnace several historical junctures and the corresponding moment needed for controller parameter correction polynary
Molten steel quality indicator measurements analyze model-free adaptive controller parameter and pseudo- partial derivative initial value sensitivity, to sensitive parameter
And pseudo- partial derivative initial value carries out off-line correction.
8. system according to claim 7, which is characterized in that the parameter correction submodel establishes module, including：
Training and test data set constructing module：By the height of historical junctures several needed for model-free adaptive controller parameter correction
Furnace body production status parametric variable and polynary molten steel quality indicator measurements of corresponding moment are divided into trained initial data and survey
Try initial data two parts；The blast-furnace body life needed for each moment training of extracting parameter correction submodel from training initial data
It produces duty parameter variable and polynary molten steel quality indicator measurements of corresponding moment forms multigroup training data, by each group training data
By being arranged to make up training dataset constantly；Extraction is accurate for test parameter correction submodel prediction from test initial data
Blast-furnace body production status parametric variable and polynary molten steel quality indicator measurements structure of corresponding moment needed for the test of each moment of rate
Into multigroup test data, by each group test data by being arranged to make up test data set constantly；
Model training module：Based on training dataset, parameter correction submodel is built using subspace state space system identification, using adding
Enter the least square method of recursion of forgetting factor, solve parameter correction submodel coefficient matrix；
Model evaluation module：Evaluation index of the root-mean-square error as parameter correction submodel is introduced, based on test data set,
If the root-mean-square error of each molten steel quality index that parameter current correction submodel estimates reaches given threshold, the ginseng
Number correction submodel is met the requirements；If the root mean square of each molten steel quality index that parameter current correction submodel estimates misses
In difference, there is any one to be not up to given threshold, then rebuild parameter correction submodel；
Model construction module：Based on test data set, using the least square method of recursion for adding in forgetting factor, parameter correction is solved
Submodel coefficient matrix obtains final parameter correction submodel.
9. system according to claim 7, which is characterized in that the control parameter sensitivity analysis module, including：
Monte Carlo simulation modules：Construct several groups and be uniformly distributed STOCHASTIC CONTROL parameter set, using parameter correction submodel as
Controlled device exports reference curve target in order to control, using accumulative root mean square tracking error as mesh with the polynary molten steel quality of setting
Offer of tender numerical value will be uniformly distributed each group control parameter in STOCHASTIC CONTROL parameter set and assign model-free adaptive controller respectively, into
Several groups of Monte Carlo experiments of row；
Sensitivity analysis module：Monte Carlo experimental results are sorted and are grouped, draw cumulative frequency curve respectively, are calculated
Curve separating degree confirms sensitive control parameter and pseudo- partial derivative initial value.
10. system according to claim 6, which is characterized in that the On-line Control unit, including：
Online data acquisition module：Current time blast-furnace body production status parametric variable is obtained in real time and obtains last moment
Polynary molten steel quality indicator measurements；
Establishing equation module：Establish the relationship between the polynary molten steel quality index of description blast furnace and blast-furnace body duty parameter variable
Linear equation；
Control module：Last moment polynary molten steel quality index output valve and the polynary molten steel quality index desired value of blast furnace are done
Difference, with new collected online blast furnace operating mode parameter feedback to model-free adaptive controller input terminal, the pseudo- local derviation of recursion update
The output of matrix number and model-free adaptive controller.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711316435.7A CN108153146B (en) | 2017-12-12 | 2017-12-12 | Model-free self-adaptive control system and method for quality of blast furnace multi-element molten iron |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711316435.7A CN108153146B (en) | 2017-12-12 | 2017-12-12 | Model-free self-adaptive control system and method for quality of blast furnace multi-element molten iron |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108153146A true CN108153146A (en) | 2018-06-12 |
| CN108153146B CN108153146B (en) | 2020-10-27 |
Family
ID=62466999
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711316435.7A Active CN108153146B (en) | 2017-12-12 | 2017-12-12 | Model-free self-adaptive control system and method for quality of blast furnace multi-element molten iron |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108153146B (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109001979A (en) * | 2018-08-24 | 2018-12-14 | 东北大学 | Blast-melted quality adaptation Robust Predictive Control method based on Lazy learning |
| CN110373510A (en) * | 2019-07-16 | 2019-10-25 | 中南大学 | A kind of narrow window control method and system of steel quality based on bimodal switching |
| CN111679584A (en) * | 2020-06-23 | 2020-09-18 | 武汉钢铁有限公司 | Regulating and controlling method and device for blast furnace smelting |
| CN111930010A (en) * | 2020-06-29 | 2020-11-13 | 华东理工大学 | LSTM network-based general MFA controller design method |
| CN112200050A (en) * | 2020-09-30 | 2021-01-08 | 同济大学 | Lake water volume space-time change analysis method and device based on multi-source satellite data |
| CN113295635A (en) * | 2021-05-27 | 2021-08-24 | 河北先河环保科技股份有限公司 | Water pollution alarm method based on dynamic update data set |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN200984678Y (en) * | 2006-12-19 | 2007-12-05 | 中国铝业股份有限公司 | Mould for soldering aluminium electrolytic cell cathode steel rod with thermite |
| US20140310228A1 (en) * | 2013-04-12 | 2014-10-16 | Yokogawa Electric Corporation | Device and method for managing plant model |
| CN104651559A (en) * | 2015-01-30 | 2015-05-27 | 东北大学 | Blast furnace liquid iron quality online forecasting system and method based on multivariable online sequential extreme learning machine |
| US20150372686A1 (en) * | 2014-06-20 | 2015-12-24 | Texas Instruments Incorporated | Microfabricated atomic clocks (mfac) & magnetometers (mfam):self-condensing silicon vapor cell cavity structure |
| CN105242545A (en) * | 2015-10-31 | 2016-01-13 | 南京南瑞继保工程技术有限公司 | Process-batch-model-based prediction method of oxygen flow rate for converter |
| CN105487385A (en) * | 2016-02-01 | 2016-04-13 | 金陵科技学院 | Internal model control method based on model free adaptive control |
| CN106249724A (en) * | 2016-09-14 | 2016-12-21 | 东北大学 | A kind of blast furnace polynary molten steel quality forecast Control Algorithm and system |
-
2017
- 2017-12-12 CN CN201711316435.7A patent/CN108153146B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN200984678Y (en) * | 2006-12-19 | 2007-12-05 | 中国铝业股份有限公司 | Mould for soldering aluminium electrolytic cell cathode steel rod with thermite |
| US20140310228A1 (en) * | 2013-04-12 | 2014-10-16 | Yokogawa Electric Corporation | Device and method for managing plant model |
| US20150372686A1 (en) * | 2014-06-20 | 2015-12-24 | Texas Instruments Incorporated | Microfabricated atomic clocks (mfac) & magnetometers (mfam):self-condensing silicon vapor cell cavity structure |
| CN104651559A (en) * | 2015-01-30 | 2015-05-27 | 东北大学 | Blast furnace liquid iron quality online forecasting system and method based on multivariable online sequential extreme learning machine |
| CN105242545A (en) * | 2015-10-31 | 2016-01-13 | 南京南瑞继保工程技术有限公司 | Process-batch-model-based prediction method of oxygen flow rate for converter |
| CN105487385A (en) * | 2016-02-01 | 2016-04-13 | 金陵科技学院 | Internal model control method based on model free adaptive control |
| CN106249724A (en) * | 2016-09-14 | 2016-12-21 | 东北大学 | A kind of blast furnace polynary molten steel quality forecast Control Algorithm and system |
Non-Patent Citations (3)
| Title |
|---|
| ZHONG-SHENG HOU 等: "《Data-Driven Model-Free Adaptive Control for a Class of MIMO Nonlinear Discrete-Time Systems》", 《IEEE TRANSACTIONS ON NEURAL NETWORKS》 * |
| ZHONGSHENG HOU 等: "《Model free adaptive control with data dropouts》", 《EXPERT SYSTEMS WITH APPLICATIONS》 * |
| 温亮 等: "《高炉炼铁过程数据驱动质量控制》", 《2017中国自动化大会》 * |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109001979A (en) * | 2018-08-24 | 2018-12-14 | 东北大学 | Blast-melted quality adaptation Robust Predictive Control method based on Lazy learning |
| CN109001979B (en) * | 2018-08-24 | 2021-05-18 | 东北大学 | Blast furnace molten iron quality self-adaptive robust prediction control method based on lazy learning |
| CN110373510A (en) * | 2019-07-16 | 2019-10-25 | 中南大学 | A kind of narrow window control method and system of steel quality based on bimodal switching |
| CN110373510B (en) * | 2019-07-16 | 2020-04-24 | 中南大学 | Molten steel quality narrow window control method and system based on bimodal switching |
| CN111679584A (en) * | 2020-06-23 | 2020-09-18 | 武汉钢铁有限公司 | Regulating and controlling method and device for blast furnace smelting |
| CN111679584B (en) * | 2020-06-23 | 2022-05-03 | 武汉钢铁有限公司 | Regulating and controlling method and device for blast furnace smelting |
| CN111930010A (en) * | 2020-06-29 | 2020-11-13 | 华东理工大学 | LSTM network-based general MFA controller design method |
| CN112200050A (en) * | 2020-09-30 | 2021-01-08 | 同济大学 | Lake water volume space-time change analysis method and device based on multi-source satellite data |
| CN113295635A (en) * | 2021-05-27 | 2021-08-24 | 河北先河环保科技股份有限公司 | Water pollution alarm method based on dynamic update data set |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108153146B (en) | 2020-10-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108153146A (en) | A kind of polynary molten steel quality MFA control system and method for blast furnace | |
| CN106249724B (en) | A kind of polynary molten steel quality forecast Control Algorithm of blast furnace and system | |
| CN111045326B (en) | Tobacco shred drying process moisture prediction control method and system based on recurrent neural network | |
| CN101286044B (en) | Coal-burning boiler system steam-temperature mixing modeling method | |
| CN104765350B (en) | Cement decomposing furnace control method and system based on Combined model forecast control technology | |
| CN101520652B (en) | Method for evaluating service reliability of numerical control equipment | |
| CN104766175A (en) | Power system abnormal data identifying and correcting method based on time series analysis | |
| CN103472723A (en) | Predictive control method and system based on multi-model generalized predictive controller | |
| CN102728624A (en) | Method for setting load distribution of finish rolling band steel | |
| CN105204333B (en) | A kind of energy consumption Forecasting Methodology for improving iron and steel enterprise's energy utilization rate | |
| CN106877746A (en) | Method for control speed and speed control unit | |
| CN103268082A (en) | Thermal error modeling method based on gray linear regression | |
| CN107390524B (en) | A kind of blast-melted quality optimization control method based on bilinearity Subspace Identification | |
| CN105159071A (en) | Method for estimating economic performance of industrial model prediction control system in iterative learning strategy | |
| CN104318090A (en) | Least square method support vector machine-based generalized prediction method in lysozyme fermentation process | |
| CN109858798B (en) | Power grid investment decision modeling method and device for correlating transformation measures with voltage indexes | |
| CN111443599A (en) | Nuclear power station steam turbine rotating speed control method for optimizing PID (proportion integration differentiation) parameters | |
| CN105013832A (en) | Hot rolled strip steel load distribution method giving consideration to rolling energy consumption and good strip shape | |
| CN111522233A (en) | Parameter self-tuning MIMO different-factor full-format model-free control method | |
| CN103853939A (en) | Combined forecasting method for monthly load of power system based on social economic factor influence | |
| CN106779384A (en) | A kind of long-term interval prediction method of steel and iron industry blast furnace gas based on Information Granularity optimum allocation | |
| CN102886384A (en) | Flatness defect identification method of 20-roller Sendzimir rolling mill based on support vector machine | |
| CN106896795A (en) | A kind of thermal power plant's varying duty coordinates Control platform evaluation method | |
| CN104267600B (en) | Ladle refining furnace Electrode Computer Control System and control method thereof | |
| CN103455003B (en) | The method for supervising produced for petrochemical industry and system |
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 |