CN107844659A - The agent model modeling method of coal water slurry gasification process - Google Patents
The agent model modeling method of coal water slurry gasification process Download PDFInfo
- Publication number
- CN107844659A CN107844659A CN201711138022.4A CN201711138022A CN107844659A CN 107844659 A CN107844659 A CN 107844659A CN 201711138022 A CN201711138022 A CN 201711138022A CN 107844659 A CN107844659 A CN 107844659A
- Authority
- CN
- China
- Prior art keywords
- coal
- model
- msup
- data
- content
- 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
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
-
- C—CHEMISTRY; METALLURGY
- C10—PETROLEUM, GAS OR COKE INDUSTRIES; TECHNICAL GASES CONTAINING CARBON MONOXIDE; FUELS; LUBRICANTS; PEAT
- C10J—PRODUCTION OF PRODUCER GAS, WATER-GAS, SYNTHESIS GAS FROM SOLID CARBONACEOUS MATERIAL, OR MIXTURES CONTAINING THESE GASES; CARBURETTING AIR OR OTHER GASES
- C10J3/00—Production of combustible gases containing carbon monoxide from solid carbonaceous fuels
- C10J3/72—Other features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/30—Circuit design
- G06F30/32—Circuit design at the digital level
- G06F30/333—Design for testability [DFT], e.g. scan chain or built-in self-test [BIST]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Chemical & Material Sciences (AREA)
- Computer Hardware Design (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Geometry (AREA)
- Biophysics (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Data Mining & Analysis (AREA)
- Computational Linguistics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Biomedical Technology (AREA)
- General Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Combustion & Propulsion (AREA)
- Oil, Petroleum & Natural Gas (AREA)
- Organic Chemistry (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Feedback Control In General (AREA)
Abstract
The present invention relates to a kind of agent model modeling method of coal water slurry gasification process, the method chooses several measurable process statuses as input variable, including oxygen coal than, the content of ashes in coal-water fluid concentration, coal slurry flow, coal in H/C elemental mole ratios, coal in O/C elemental mole ratios and coal, simultaneously, several measurable process statuses are chosen as target output variable, including CO content, CO in exiting syngas2Content, H2Content, charcoal percent conversion in the temperature in exit and coal.After being sampled using Latin Hypercube Sampling method to input variable, input data is analyzed and handled.The data model established using Kriging agent models between input variable and output variable, optimal agent model parameter is gone out by the particle swarm optimization algorithm after improvement.The fitting precision of the model is high, tracking effect is good, model generalization ability is strong, has preferable industry park plan directive significance.
Description
Technical field
The invention belongs in energy industry and field of chemical engineering, be related to a kind of agent model of coal water slurry gasification process to build
The method of mould method, the especially agent model of the operating parameter of coal water slurry gasification course of reaction and gasification product modeling.
Background technology
Not only utilization rate is low but also very big pollution is caused to environment for traditional coal utilization mode, in order to realize ecology
Sustainable development, coal gasification technology have become accelerate Coal Industry Development primary study content.Coal gasification industry has
There are scale and maximization feature, serve primarily in coal base chemical field and combined generating system.By coal and gasifying agent in height
The a series of complex chemical reaction occurred under warm hyperbaric environment, the process for producing clean energy resource are referred to as coal gasification technology, its core
Heart technology includes：Static bed coal gasification technology, In The Fluidized Bed Coal Gasification Technology and air flow bed Coal Gasification Technology.For gasification furnace
The optimization problem of device is always the hot issue of domestic and international gasification technology research, and the modelling of gasification furnace is matched somebody with somebody to optimizing raw material
Put, improve synthesis gas yield, reduce production cost and play an important roll, but at present on coal gasification course Modeling Research, it is domestic
The report of outer pertinent literature is also relatively fewer.
Because GE coal water slurry gasification technologies have the characteristics that coal adaptability is strong, effective gas high income and drawn very early by China
Enter, have become the mainstream technology of China's Coal Gasification Technology now.The gasification technology is divided into according to the mode of recuperation of heat
Three kinds of technological processes：Gasification flow, the pot destroying process with cooler and the useless pot Quench combined process flow of direct chilling-type.Wherein
Direct chilling-type GE gasification furnaces are almost applied to all water-coal-slurry chemical industry synthesis production processes, generally comprise coal slurry and prepare system
System, gasified boiler system, crude synthesis gas cleaning system, slag recovery and grey water treatment system, are fed using water-coal-slurry, in air flow bed
Pressurized gasification, reaction generation synthesis gas.
Coal water slurry gasification process is the complex chemical reaction occurred in the environment of HTHP, and its model has non-thread
The features such as property, multiple coupled, multivariable, multiple constraint.Coal water slurry gasifying device gasification is studied, to steady running condition, is improved
Operational efficiency and coal utilizaton rate have very important significance.The present invention is on the basis of gasification furnace Analysis on Mechanism, with GE water
Coal slurry gasification furnace is object, by entering with Kriging modeling methods to the gasification furnace furnace temperature, effective gas component and efficiency of carbon con version
Row modeling and prediction.
The content of the invention
It is an object of the invention to provide a kind of agent model that can replace coal slurry gasifier mechanism model, Latin is used
The hypercube method of sampling is to input variable X=[foc, fcon, fflow, fO/C, fH/C, fash] sampled, obtain corresponding output and become
AmountThe obtained sample point of sampling is normalized and standardization after, for pair
Kriging agent models are trained, and by the particle swarm optimization algorithm correlation model parameters after improvement, and then are obtained full
The agent model optimized parameter required enough, output data can be predicted according to the input condition of coal slurry gasifier.
The method of the agent model of present invention structure coal water slurry gasification course of reaction includes：Choose several measurable processes
State (including oxygen coal than, in coal-water fluid concentration, coal slurry flow, coal in H/C elemental mole ratios, coal in O/C elemental mole ratios and coal
Content of ashes) input variable is used as, choose several measurable process statuses (including CO content, CO in exiting syngas2's
Content, H2Content, charcoal percent conversion in the temperature in exit and coal) be used as target output variable, utilize Latin hypercube
After the method for sampling samples to input variable, input data is analyzed and handled, utilize Kriging agent models afterwards
The data model established between input variable and output variable, optimal agency is gone out by the particle swarm optimization algorithm after improvement
Model parameter.
In one or more embodiments, methods described comprises the following steps：
Step 1：Actual conditions are run according to commercial plant, determine input variable X=[foc, fcon, fflow, fO/C, fH/C,
fash] scope, and input variable is sampled using Latin Hypercube Sampling method, passes through coal water slurry gasification process mechanism
Model obtains output variable corresponding to sampled pointWherein, focFor oxygen coal ratio, fconFor coal slurry
Concentration, fflowFor coal slurry flow, fH/CFor the H/C mol ratios of coal, fO/CFor the O/C mol ratios of coal, fashFor pit ash content,
yCO、yCO2、yH2、yTAnd yCCCO contents, CO respectively in exiting syngas2Content, H2Content, the temperature in exit and coal
Middle charcoal percent conversion；
Step 2：The sample set that step 1 obtains is normalized, the sample data of random taking-up 9/10 is as training set
S1, for training and modeling；Test set S2 of remaining 1/10 sample data as model, verified for model；
Step 3：To training set S1, the average and variance of input variable and output variable are obtained respectively, is standardized place
Reason；
Step 4：Start Kriging models, initialization model parameter, the data input Kriging that step 3 processing is obtained
Model；
Step 5：Using the optimized parameter of PSO Algorithm correlation function, and then obtain the regression parameter of model and pre-
Survey error；With
Step 6：The accuracy of model prediction is assessed, if the average relative variance and average relative error of future position
Meet industrial required precision, then model training success, otherwise chooses sample point again.
In one or more embodiments, the scope of input variable is
foc∈ [0.96,1.06],
fcon=[0.57,0.63],
fflow∈ [21500,22500], its unit are kg/h,
fH/C∈ [0.8,0.9],
fO/C∈ [0.10,0.13],
fash∈ [0.07,0.075].
In one or more embodiments, the method for sampling comprises the steps of in step 2：
A, sample：System input vector X=[foc, fcon, fflow, fO/C, fH/C, fash], sampling sum is N, it is assumed that its point
Cloth function is Gi=Fi(Xi), corresponding to input variable X value, digital average of the distribution function on the longitudinal axis is divided into N etc.
Point, the value that a numerical point is used as distribution function is arbitrarily chosen in caused N number of nonoverlapping section by random function
GI, n, the sampled value of selected numerical point is obtained by inverse function, i.e. n-th of sampled value is XI, n=Fi -1(GI, n), wherein Fi -1For Fi
Inverse function, the sampled value of the stochastic variable of acquisition ultimately forms i × N sample matrix；
B, sort：The space progress of decile is randomly ordered, then carry out stochastical sampling in each section.
In one or more embodiments, in step 2, after sampling, line number is entered to the training sample of acquisition respectively
Data preprocess, rejecting abnormalities data point, corresponding qualified data are obtained, afterwards data are normalized with operation, it is normalized
Scope is [- 1 1], and normalization formula is as follows：
Y=(ymax-ymin)×(x-xmin)/(xmax-xmin)+ymin
Y in formulamin=-1, ymax=1.
In one or more embodiments, the standardization formula in step 3 to sample data is as follows：
MX=mean (X)；SX=std (X)；
Forj=1：N, X (：, j)=(X (：, j) and-mX (j))/sX (j)；end
MY=mean (Y)；SY=std (Y)；
Forj=1：Q, Y (：, j)=(Y (：, j) and-mY (j))/sY (j)；end
N represents the dimension of input vector in formula, and q represents the dimension of output vector.
In one or more embodiments, in step 4, regression model is a rank multinomial, and correlation model is Gaussian function
Number.
In one or more embodiments, the particle cluster algorithm employed in step 5 introduces linear inertia power to be a kind of
The improvement particle cluster algorithm of weight, the algorithm is a kind of random search algorithm based on population.
In one or more embodiments, step 6 solves following formula optimization problem using intelligent optimization algorithm：
In one or more embodiments, the average relative variance of industrial required precision future position and average in step 6
Relative error is less than 5%.
The present invention also provides a kind of method for optimizing water-gas gasification reaction process, and methods described includes：
(1) using Latin Hypercube Sampling method to input variable X=[foc, fcon, fflow, fO/C, fH/C, fash] adopted
Sample, obtain output variable corresponding to sampled pointWherein, focFor oxygen coal ratio, fconIt is dense for coal slurry
Degree, fflowFor coal slurry flow, fH/CFor the H/C mol ratios of coal, fO/CFor the O/C mol ratios of coal, fashFor pit ash content, yCO、
yCO2、yH2、yTAnd yCCCO contents, CO respectively in exiting syngas2Content, H2Carbon in content, the temperature in exit and coal
Conversion ratio；
(2) data obtained for step (1) are normalized；
(3) by the data after normalized in step (2), average and the side of input variable and output variable are obtained respectively
Difference, it is standardized；
(4) side of the data input any one of according to claim 1-9 after step (3) is standardized
Method builds obtained agent model, obtains corresponding output result；With
(5) course of reaction is optimized according to the output result；
In certain embodiments, the optimization method also includes the step that agent model is built according to method described herein
Suddenly.
The present invention can correctly describe influence of the coal slurry gasifier performance variable change to output result.Start the agency
Model, by training set data after pretreatment, agent model is trained as the input of model, the data that will be predicted are passed through
This agent model is inputted after pretreatment, agent model can predict the corresponding output result of gasification furnace, instruct gasification reaction mistake
The optimization operation operation of journey.
Brief description of the drawings
Fig. 1 is agent model frame diagram.
Fig. 2 is particle cluster algorithm iterative process figure.
Fig. 3 is Kriging agent model modeling process schematic diagrames.
Fig. 4 is the result being predicted using the agent model constructed by embodiment.
Embodiment
Fig. 1 shows the frame diagram of coal water slurry gasification process agent model.With oxygen coal ratio, coal-water fluid concentration, coal slurry flow, coal
One or more of content of ashes in middle H/C elemental mole ratios, coal in O/C elemental mole ratios and coal is used as input variable,
With CO content, CO in exiting syngas2Content, H2Content, in the temperature in exit and coal in charcoal percent conversion one
Individual or multiple conduct target output variables.
Fig. 2 is particle cluster algorithm iterative process figure.Fig. 3 shows Kriging agent model modeling process schematic diagrames, including：
1st, Selection Model input variable and output variable；
2nd, process mechanism model is sampled using Latin Hypercube Sampling method；
3rd, the normalized of sample；
4th, the standardization of sample；
5th, training system network；
6th, agent model is built；With
7th, the accuracy of model prediction is assessed, if the average relative variance and average relative error of future position meet
Industrial required precision, then model training success, otherwise chooses sample point again.
These steps will be hereafter described in detail.It should be understood that within the scope of the present invention, above-mentioned each skill of the invention
It can be combined with each other between art feature and each technical characteristic specifically described in below (eg embodiment), so as to form preferably
Technical scheme.
First, the selection of mode input variable and output variable
Actual conditions can be run according to commercial plant, choose several measurable process statuses as input variable, choose
Several measurable process statuses are as target output variable.
Generally, optional take makees the process status of input variable and compares f including oxygen coaloc, coal-water fluid concentration fcon, coal slurry flow
fflow, coal H/C mol ratios fH/C, coal O/C mol ratios fO/CAnd pit ash content fashIn one or more, be preferentially
All.The process status as target output variable that can be chosen includes the content y of CO in exiting syngasCO、CO2Content
yCO2、H2Content yH2, exit temperature yTWith charcoal percent conversion y in coalCCIn one or more, be preferentially all.
2nd, process mechanism model is sampled using Latin Hypercube Sampling method
This step determines input variable and its scope first, and then input variable is entered using Latin Hypercube Sampling method
Row sampling.
In certain embodiments, the scope of input variable can be set to：
foc∈ [0.96,1.06], fcon=[0.57,0.63], fflow∈ [21500,22500], its unit are kg/h, fH/C
∈ [0.8,0.9], fO/C∈ [0.10,0.13], fash∈ [0.07,0.075].
It can be sampled using Latin Hypercube Sampling method well known in the art.Pass through coal water slurry gasification process mechanism mould
Type obtains output variable corresponding to sampled point.
In certain embodiments, sampling comprises the following steps：
A, sample：System input vector X=[foc, fcon, fflow, fO/C, fH/C, fash], sampling sum is N
Assuming that its distribution function is Gi=Fi(Xi), corresponding to input variable X value, by distribution function on the longitudinal axis
Digital average is divided into N deciles, and arbitrarily choosing a numerical point in caused N number of nonoverlapping section by random function is used as
The value G of distribution functionI, n, the sampled value of selected numerical point is obtained by inverse function, i.e. n-th of sampled value is XI, n=Fi -1
(GI, n), wherein Fi -1For FiInverse function, the sampled value of the stochastic variable of acquisition ultimately forms i × N sample matrix；
B, sort：Because the computational accuracy of sampled data is by the interdependence effects between different sampled values, therefore, first will
The space progress of decile is randomly ordered, then carries out stochastical sampling in each section.
3rd, the normalized of sample
After sampling, the training sample (input variable and output variable that i.e. step 2 obtains) of acquisition is carried out respectively
Data prediction.With reference to actual industrial background knowledge, part exceptional data point is rejected.It is corresponding by being obtained after above-mentioned pretreatment
Qualified data.In order to place data into unified scope to facilitate data processing, while preferably accelerate the convergence of network, logarithm
Operated according to being normalized.Normalization scope can be set to [- 1 1], and normalization formula is as follows：
Y=(ymax-ymin)×(x-xmin)/(xmax-xmin)+ymin
Y in formulamin=-1, ymax=1.
Generally, 9/10 sample data is taken at random to the training sample of acquisition as training set S1, for training and modeling；
Test set S2 of remaining 1/10 sample data as model, for predicting and verifying.
4th, the standardization of sample
In this step, the sample data obtained to step 3 is standardized, and standardization formula is as follows：
MX=mean (X)；SX=std (X)；
Forj=1：N, X (：, j)=(X (：, j) and-mX (j))/sX (j)；end
MY=mean (Y)；SY=std (Y)；
Forj=1：Q, Y (：, j)=(Y (：, j) and-mY (j))/sY (j)；end
N represents the dimension of input vector in formula, and q represents the dimension of output vector.
5th, training system network
The step for include starting Kriging models, initialize network architecture parameters θ, and step 4 handled what is obtained
The data input Kriging models.
θ is sample point in all directions constant parameter.θ initial values are arranged to 10, and scope is arranged to [0.1,20].Calculate every two
The distance between individual sample point R, m=792, it is as follows to be expressed as matrix form：
X is training sample point in formula.
In this step, regression model and correlation model are used.
Conventional regression function has constant type, lienar for and quadratic form, as follows respectively：
Constant type：P=1：f1(x)=1
Lienar for：P=n+1：f1(x)=1, f2(x)=x1... ..fn+1(x)=xn
Quadratic form：
f2(x)=x1... fn+1(x)=xn
In certain embodiments, selection uses lienar for regression function, therefore
Correlation function, its model can be represented by formula below：
Wherein djIt is expressed as distance between tested point and testing site, dj=wj-xj, j=1 ..., n, xjIt is x in j-th of side
To coordinate, vjFor coordinates of the v in j-th of direction, θjFor constant parameter of the experimental data point in j-th of direction.Rj(θj, dj) be
The kernel function of correlation function.
Kernel function in correlation function generally has following several：
EXP model exp (- θj|dj|)
EXPG models0 ＜ θn+1≤2
Gauss model (GAUSS)
Linear model (LIN) max { 0,1- θj|dj|}
Spherical model (SPHERICAL)
Cubic model (CUBIC)
Spline function model (SPLINE)
Correlation function is ordinarily selected to exponential model, spherical model, Gauss model, is satisfied by spatial coherence and increases with distance
Big and be gradually reduced, the shape of the correlation function of spherical model and exponential model near origin is linear, and Gauss model
The curve of correlation function parabolically shape near origin, has more preferable continuity.
In certain embodiments, present invention selection Gaussian function correlation model, its fitting result are preferable.
6th, agent model is built
The response of system is described as following expression with independent variable, including returns part and nonparametric part (i.e. herein
Described agent model)：
Y=F β+z (x)
Wherein y is system output vector, and F is regression function matrix, and β is regression parameter, and z (x) is random error, there is provided right
The approximation of partial deviations, there is following statistical property：
E [z (x)]=0
Var [z (x)]=σ2
Cov[Z(xi), Z (xj)]=σ2[Rij(θ, xi, xj)]
Its generalized linear least-squares estimation is：
β*=(FTR-1F)-1FTR-1Y
Associated vector between unknown point x and m known sampled points is：
R (x)=[R (x, x1), R (x, x2) ..., R (x, xm)]T,
Unknown point x estimate expression formula is：
Wherein：Rγ*=Y-F β*
Assuming that the identical i.e. R=I of variance between uncorrelated between error z (x) and error, then β*Least square solution
For：
(FTF)β*=FTY
Corresponding variance evaluation is：
Agent model random error z (x) described herein meets Var [z (x)]=σ2, Cov [Z (xi), Z (xj)]=σ2[Rj
(θ, xi, xj)], make distance matrix R be expressed as R=WWTThen there is W-1Y=W-1Fβ+W-1z(x)；
Order
It is concluded that
F, Y are respectively by W-1F, W-1Y substitutes, then β*And σ2It is represented as：
(FTR-1F)β*=FTR-1Y
Using maximum-likelihood method come optimize determine θ when, when correlation function is Gaussian process, can by solve following formula come
Correlation model parameters θ in calculation procedure five
The optimal solution of above formula is obtained by particle cluster algorithm optimizing, the iterative formula of particle cluster algorithm is as follows：
K is current iteration number；For example speed；c1And c2For acceleration factor, r1And r2It is distributed across [0,1] area
Between random number；ω is inertia weight；ωstartFor initial inertia weight,；ωendInertia weight during to iterate to maximum times,；
Tmax genFor maximum iteration.The scope of the inertia weight is [0,1].Iterations is not specifically limited, generally
Tens to hundreds of generations；Population scale is also not specifically limited, generally tens to hundreds of.
7th, the accuracy of model prediction is assessed
The relative error formula of average relative variance peace of future position for being mentioned in step 6 is as follows：
Wherein a is test sample point number, and y (x) is prediction desired value,For predicted value.
Present invention additionally comprises the method using the agent model optimization water-gas gasification reaction process obtained constructed by this paper,
Methods described includes：
(1) using Latin Hypercube Sampling method to input variable X=[foc, fcon, fflow, fO/C, fH/C, fash] adopted
Sample, obtain output variable corresponding to sampled pointWherein, focFor oxygen coal ratio, fconIt is dense for coal slurry
Degree, fflowFor coal slurry flow, fH/CFor the H/C mol ratios of coal, fO/CFor the O/C mol ratios of coal, fashFor pit ash content, yCO、
yCO2、yH2、yTAnd yCCCO contents, CO respectively in exiting syngas2Content, H2Carbon in content, the temperature in exit and coal
Conversion ratio；
(2) data obtained for step (1) are normalized；
(3) by the data after normalized in step (2), average and the side of input variable and output variable are obtained respectively
Difference, it is standardized；
(4) agent model constructed by data input this paper after step (3) is standardized, obtain corresponding
Output result；With
(5) course of reaction is optimized according to the output result.
Hereafter the present invention is specifically described by embodiment.It is necessarily pointed out that following examples are only used
In the invention will be further described, it is impossible to be interpreted as limiting the scope of the invention, professional and technical personnel in the field
Some the nonessential modifications and adaptations made according to present disclosure, still fall within protection scope of the present invention.
Embodiment
The present embodiment chooses Latin Hypercube Sampling method and input data is carried out according to actual industrial data variation scope
Sampling.Randomly select 792 groups of training sample points and 88 groups of test sample points.
In order to eliminate influence of the dimension to sample data, sample point is normalized, training set data is carried out
Standardization.Initialize network architecture parameters θ, θ is sample point in all directions constant parameter, is used in steps of 5.At the beginning of θ
Value is arranged to 10, and scope is arranged to [0.1,20].
The distance between each two sample point R, m=792 are calculated, it is as follows to be expressed as matrix form：
X is training sample point in formula.
Lienar for regression function is selected, wherein n=6 is the dimension of input sample, therefore
Select the preferable Gaussian function mutation model of fitting result.Because it is both positive definite and symmetrical that R, which is, for letter
Change and calculate, Cholesky decomposition can be carried out to R：
R=CCT
C is the Qiao Lisi factors in formula
Separately Then have
In order to prevent ill R, to matrixCarry out QR decomposition
It can thus be concluded thatSolve expression formula：
And variance evaluation value expression：
Using maximum-likelihood method come optimize determine θ when, when correlation function is Gaussian process, can by solve following formula come
Calculate correlation model parameters θ
The optimal solution of above formula is obtained by particle cluster algorithm optimizing, the iterative formula of particle cluster algorithm is as follows：
ω (k)=ωstart-(ωstart-ωend)×k/Tmax gen
K is current iteration number；For example speed；c1And c2For acceleration factor, 1.495 are arranged to；ω is inertia
Weight；ωstartFor initial inertia weight, 0.9 is arranged to；ωendInertia weight during to iterate to maximum times, is arranged to 0.4；
Tmax genFor maximum iteration, it was arranged to for 50 generations, population scale is arranged to 20.
It is as follows for the relative error formula of average relative variance peace of future position：
Wherein a is test sample point number, and y (x) is prediction desired value,For predicted value.
The average of its average relative variance and average relative error after running 10 times is as shown in the table：
The value of average relative variance and average relative error meets model accuracy requirement (＜ 5%).
To sum up, the agent model of coal slurry gasifier input and output can be obtained.By service data after data processing
As the input of model, the output data of gasification furnace can be predicted, the optimization operation to gasification is with certain guidance work
With.
Claims (10)
1. a kind of method for the agent model for building coal water slurry gasification course of reaction, methods described include：Choose oxygen coal ratio, coal slurry
Content of ashes in concentration, coal slurry flow, coal in H/C elemental mole ratios, coal in O/C elemental mole ratios and coal becomes as input
Amount, choose CO content, CO in exiting syngas2Content, H2Content, charcoal percent conversion in the temperature in exit and coal
As target output variable, input variable is sampled using Latin Hypercube Sampling method, input data is analyzed
With processing, the data model established afterwards using Kriging agent models between input variable and output variable, after improvement
Particle swarm optimization algorithm go out optimal agent model parameter, so as to build to obtain the agent model.
2. the method as described in claim 1, it is characterised in that methods described comprises the following steps：
Step 1：Actual conditions are run according to commercial plant, determine input variable X=[foc, fcon, fflow, fO/C, fH/C, fash]
Scope, and input variable is sampled using Latin Hypercube Sampling method, obtained by coal water slurry gasification process mechanism model
Take output variable corresponding to sampled pointWherein, focFor oxygen coal ratio, fconFor coal-water fluid concentration,
fflowFor coal slurry flow, fH/CFor the H/C mol ratios of coal, fO/CFor the O/C mol ratios of coal, fashFor pit ash content, yCO、
yCO2、yH2、yTAnd yCCCO contents, CO respectively in exiting syngas2Content, H2Carbon in content, the temperature in exit and coal
Conversion ratio；
Step 2：The sample set that step 1 obtains is normalized；
Step 3：The data through normalized obtained to step 2, obtain respectively input variable and output variable average and
Variance, it is standardized；
Step 4：Start Kriging models, initialization model parameter, the data input that step 3 standardization is obtained
Kriging models；
Step 5：Using the optimized parameter of PSO Algorithm correlation function, and then obtain the regression parameter and prediction mistake of model
Difference, so as to build to obtain the agent model, its expression formula is as follows：
Y=F β+z (x)
Wherein y is system output vector, and F is regression function matrix, and β is regression parameter, and z (x) is random error.
3. method as claimed in claim 1 or 2, it is characterised in that the scope of the input variable is
fcon=[0.57,0.63]；fflow∈ [21500,22500], its unit are kg/h；fH/C∈ [0.8,0.9]；fO/C∈ [0.10,
0.13]；With
4. method as claimed in claim 1 or 2, it is characterised in that in step 2, after sampling, respectively to the sample of acquisition
Data prediction is carried out, rejecting abnormalities data point, corresponding qualified data is obtained, afterwards data is normalized with operation, its
It is [- 1 1] to normalize scope, and normalization formula is as follows：
Y=(ymax-ymin)×(x-xmin)/(xmax-xmin)+ymin
Y in formulamin=-1, ymax=1.
5. method as claimed in claim 1 or 2, it is characterised in that the standardization formula in step 3 to sample data is such as
Under：
MX=mean (X)；SX=std (X)；
Forj=1：N, X (：, j)=(X (：, j) and-mX (j))/sX (j)；end
MY=mean (Y)；SY=std (Y)；
Forj=1：Q, Y (：, j)=(Y (：, j) and-mY (j))/sY (j)；end
N represents the dimension of input vector in formula, and q represents the dimension of output vector.
6. method as claimed in claim 2, it is characterised in that step 4 includes：Start Kriging models, initialize network knot
Structure parameter θ, and the data input Kriging models that step 3 processing is obtained；
Wherein, θ be sample point in all directions constant parameter, θ initial values are arranged to 10, and scope is arranged to [0.1,20], calculate every
The distance between two sample points R, m=792, it is as follows to be expressed as matrix form：
X is training sample point in formula；
Preferably, the regression model used in step 4 elects a rank multinomial as, and correlation model is Gaussian function；
Preferably, the function of regression model is lienar for regression function, therefore
7. method as claimed in claim 2, it is characterised in that
Random error z (x) personality presentation is following formula：
E [z (x)]=0
Var [z (x)]=σ2
Cov[Z(xi), Z (xj)]=σ2[Rij(θ, xi, xj)]；
Assuming that the identical i.e. R=I of variance between uncorrelated between error z (x) and z (x), then β*Least square solution be：
(FTF)β*=FTY
Corresponding variance evaluation is：
<mrow>
<msup>
<mi>&sigma;</mi>
<mn>2</mn>
</msup>
<mo>=</mo>
<mfrac>
<mn>1</mn>
<mi>m</mi>
</mfrac>
<msup>
<mrow>
<mo>(</mo>
<mi>Y</mi>
<mo>-</mo>
<msup>
<mi>F&beta;</mi>
<mo>*</mo>
</msup>
<mo>)</mo>
</mrow>
<mi>T</mi>
</msup>
<mrow>
<mo>(</mo>
<mi>Y</mi>
<mo>-</mo>
<msup>
<mi>F&beta;</mi>
<mo>*</mo>
</msup>
<mo>)</mo>
</mrow>
</mrow>
And random error z (x) meets Var [z (x)]=σ2, Cov [Z (xi), Z (xj)]=σ2[Rj(θ, xi, xj)], order matrix R tables
It is shown as R=WWTThen there is W-1Y=W-1Fβ+W-1z(x)；
Order
It is concluded that
F, Y are respectively by W-1F, W-1Y substitutes, then β*AndVariance evaluation σ2It is represented as：
(FTR-1F)β*=FTR-1Y
<mrow>
<msup>
<mi>&sigma;</mi>
<mn>2</mn>
</msup>
<mo>=</mo>
<mfrac>
<mn>1</mn>
<mi>m</mi>
</mfrac>
<msup>
<mrow>
<mo>(</mo>
<mi>Y</mi>
<mo>-</mo>
<msup>
<mi>F&beta;</mi>
<mo>*</mo>
</msup>
<mo>)</mo>
</mrow>
<mi>T</mi>
</msup>
<msup>
<mi>R</mi>
<mrow>
<mo>-</mo>
<mn>1</mn>
</mrow>
</msup>
<mrow>
<mo>(</mo>
<mi>Y</mi>
<mo>-</mo>
<msup>
<mi>F&beta;</mi>
<mo>*</mo>
</msup>
<mo>)</mo>
</mrow>
<mo>.</mo>
</mrow>
8. the method as described in right 1 or 2, it is characterised in that following formula optimization problem is solved using intelligent optimization algorithm：
9. the method as described in right 1 or 2, it is characterised in that methods described also includes：The accuracy of model prediction is commented
Estimate, if the average relative variance and average relative error of future position meet industrial required precision, model training success, otherwise weigh
It is new to choose sample point；
Preferably, the average relative variance and average relative error of industrial required precision future position are less than 5%.
10. a kind of method for optimizing water-gas gasification reaction process, methods described include：
(1) using Latin Hypercube Sampling method to input variable
X=[foc, fcon, fflow, fO/C, fH/C, fash] sampled, obtain output variable corresponding to sampled pointWherein, focFor oxygen coal ratio, fconFor coal-water fluid concentration, fflowFor coal slurry flow, fH/CFor coal
H/C mol ratios, fO/CFor the O/C mol ratios of coal, fashFor pit ash content, yCO、yCO2、yH2、yTAnd yCCRespectively outlet is closed
Into the CO contents in gas, CO2Content, H2Charcoal percent conversion in content, the temperature in exit and coal；
(2) data obtained for step (1) are normalized；
(3) by the data after normalized in step (2), the average and variance of input variable and output variable are obtained respectively,
It is standardized；
(4) data input after step (3) is standardized is according to the method structure any one of claim 1-9
Obtained agent model is built, obtains corresponding output result；With
(5) course of reaction is optimized according to the output result；
Preferably, the step of methods described also includes building the agent model according to the method described in claim 1-9.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711138022.4A CN107844659B (en) | 2017-11-16 | 2017-11-16 | Agent model modeling method for coal water slurry gasification process |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711138022.4A CN107844659B (en) | 2017-11-16 | 2017-11-16 | Agent model modeling method for coal water slurry gasification process |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107844659A true CN107844659A (en) | 2018-03-27 |
| CN107844659B CN107844659B (en) | 2021-05-14 |
Family
ID=61679612
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711138022.4A Active CN107844659B (en) | 2017-11-16 | 2017-11-16 | Agent model modeling method for coal water slurry gasification process |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107844659B (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2020062806A1 (en) * | 2018-09-28 | 2020-04-02 | 东南大学 | Improved ina feedforward control method for post-combustion co2 capture system |
| CN111292811A (en) * | 2020-01-22 | 2020-06-16 | 华东理工大学 | Product prediction method and system for aromatic disproportionation production link |
| CN111312338A (en) * | 2020-02-10 | 2020-06-19 | 华东理工大学 | Product prediction method and system for aromatic hydrocarbon isomerization production link |
| CN112287484A (en) * | 2020-10-29 | 2021-01-29 | 复旦大学 | Complex engineering system reliability design method based on vector agent model |
| CN115130246A (en) * | 2022-07-05 | 2022-09-30 | 北京理工大学 | Parametric design method for diameters of intake and exhaust valves of diesel engines with different strengthening degrees |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102096375A (en) * | 2011-02-25 | 2011-06-15 | 华东理工大学 | Soft measurement and optimization control method for chlorine concentration in black water of coal water slurry gasifying device |
| CN102175345A (en) * | 2011-01-06 | 2011-09-07 | 华东理工大学 | Soft measurement method for fire box temperature of multi-nozzle opposed coal water slurry gasification furnace |
| CN102184452A (en) * | 2011-03-04 | 2011-09-14 | 华东理工大学 | Soft measurement method for component gas of 'Texaco synthesis gas' |
| CN103336869A (en) * | 2013-07-05 | 2013-10-02 | 广西大学 | Multi-objective optimization method based on Gaussian process simultaneous MIMO model |
| US20140173537A1 (en) * | 2012-07-11 | 2014-06-19 | University Of North Texas | Methodology for nanoscale technology based mixed-signal system design |
| CN106777922A (en) * | 2016-11-30 | 2017-05-31 | 华东理工大学 | A kind of CTA hydrofinishings production process agent model modeling method |
-
2017
- 2017-11-16 CN CN201711138022.4A patent/CN107844659B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102175345A (en) * | 2011-01-06 | 2011-09-07 | 华东理工大学 | Soft measurement method for fire box temperature of multi-nozzle opposed coal water slurry gasification furnace |
| CN102096375A (en) * | 2011-02-25 | 2011-06-15 | 华东理工大学 | Soft measurement and optimization control method for chlorine concentration in black water of coal water slurry gasifying device |
| CN102184452A (en) * | 2011-03-04 | 2011-09-14 | 华东理工大学 | Soft measurement method for component gas of 'Texaco synthesis gas' |
| US20140173537A1 (en) * | 2012-07-11 | 2014-06-19 | University Of North Texas | Methodology for nanoscale technology based mixed-signal system design |
| CN103336869A (en) * | 2013-07-05 | 2013-10-02 | 广西大学 | Multi-objective optimization method based on Gaussian process simultaneous MIMO model |
| CN106777922A (en) * | 2016-11-30 | 2017-05-31 | 华东理工大学 | A kind of CTA hydrofinishings production process agent model modeling method |
Non-Patent Citations (2)
| Title |
|---|
| 王晓强 等: "基于kriging的差分进化算法及其在苯乙烯流程优化中的应用", 《化工学报》 * |
| 祁荣宾 等: "基于kriging代理模型的PX氧化过程优化", 《华东理工大学学报（自然科学版）》 * |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2020062806A1 (en) * | 2018-09-28 | 2020-04-02 | 东南大学 | Improved ina feedforward control method for post-combustion co2 capture system |
| CN111292811A (en) * | 2020-01-22 | 2020-06-16 | 华东理工大学 | Product prediction method and system for aromatic disproportionation production link |
| CN111312338A (en) * | 2020-02-10 | 2020-06-19 | 华东理工大学 | Product prediction method and system for aromatic hydrocarbon isomerization production link |
| WO2021159822A1 (en) * | 2020-02-10 | 2021-08-19 | 华东理工大学 | Method and system for predicting product of aromatic hydrocarbon isomerization production chain |
| CN112287484A (en) * | 2020-10-29 | 2021-01-29 | 复旦大学 | Complex engineering system reliability design method based on vector agent model |
| CN112287484B (en) * | 2020-10-29 | 2021-12-07 | 复旦大学 | Complex engineering system reliability design method based on vector agent model |
| CN115130246A (en) * | 2022-07-05 | 2022-09-30 | 北京理工大学 | Parametric design method for diameters of intake and exhaust valves of diesel engines with different strengthening degrees |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107844659B (en) | 2021-05-14 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107844659A (en) | The agent model modeling method of coal water slurry gasification process | |
| Kargbo et al. | Optimisation of two-stage biomass gasification for hydrogen production via artificial neural network | |
| Zhao et al. | A data-enhanced distributionally robust optimization method for economic dispatch of integrated electricity and natural gas systems with wind uncertainty | |
| CN104651559B (en) | Blast furnace liquid iron quality online forecasting system and method based on multivariable online sequential extreme learning machine | |
| Liu et al. | Data-driven modeling for UGI gasification processes via an enhanced genetic BP neural network with link switches | |
| Yin et al. | An integrated computational intelligence technique based operating parameters optimization scheme for quality improvement oriented process-manufacturing system | |
| Li et al. | Modeling of ash agglomerating fluidized bed gasifier using back propagation neural network based on particle swarm optimization | |
| Wang et al. | Operation optimization of Shell coal gasification process based on convolutional neural network models | |
| Sen et al. | A multi-objective optimization approach to optimal sensor location problem in IGCC power plants | |
| Yang et al. | A hybrid data-driven machine learning framework for predicting the performance of coal and biomass gasification processes | |
| CN102399594A (en) | Method for optimizing operating working condition of 'Texaco gasifier' on line | |
| Gong et al. | Energy efficiency enhancement of energy and materials for ethylene production based on two-stage coordinated optimization scheme | |
| Stanger et al. | Dynamic modeling of dual fluidized bed steam gasification for control design | |
| Ma et al. | GRU-Auto-Encoder neural network based methods for diagnosing abnormal operating conditions of steam drums in coal gasification plants | |
| Liu et al. | Temporal hypergraph attention network for silicon content prediction in blast furnace | |
| Zhang et al. | Grouped time series networks for grade monitoring of zinc tailings with multisource features | |
| Tao et al. | Optimization Analysis of Power Coal-Blending Model and Its Control System Based on Intelligent Sensor Network and Genetic Algorithm. | |
| CN115762664A (en) | Method for predicting gas production component in high-temperature gasification process of multi-component organic waste | |
| Ji et al. | Prediction of syngas compositions in shell coal gasification process via dynamic soft-sensing method | |
| Zhang et al. | Monitoring method for gasification process instability using BEE-RBFNN pattern recognition | |
| Yang et al. | A multi-objective optimization model based on long short-term memory and non-dominated sorting genetic algorithm II | |
| Cao et al. | A novel 3-layer mixed cultural evolutionary optimization framework for optimal operation of syngas production in a Texaco coal-water slurry gasifier | |
| CN114897435A (en) | One-stage furnace operation abnormity detection method based on instant neighbor reconstruction error | |
| Liu et al. | Data–driven online modelling for a UGI gasification process using modified lazy learning with a relevance vector machine | |
| Li et al. | Two-stage multi-level equipment grey state prediction model and application |
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 |