Summary of the invention
The object of the invention is to overcome the deficiencies in the prior art, a kind of gas purifying process modeling optimization method based on Unscented kalman neural network is provided, it can set up accurately high sulfur-containing natural gas cleaning and desulfurization industrial process model reliably, finished product gas output can be improved based on this, reducing sweetening process energy consumption, having important practical significance to instructing actual industrial production.
The object of the present invention is achieved like this:
Based on a gas purifying process modeling optimization method for Unscented kalman neural network, it is characterized in that the method is carried out as follows:
Step 1: the input variable determining high sulfur-containing natural gas cleaning and desulfurization model of process: select m the process operation parameter that can be effectively controlled in high sulfur-containing natural gas cleaning and desulfurization art production process as mode input variable, wherein, m=10, input variable is respectively: x
1represent desulfuration absorbing tower amine liquid inlet flow rate, x
2represent tail gas absorber amine liquid inlet flow rate, x
3represent unstripped gas treatment capacity, x
4represent half rich amine solution internal circulating load, x
5represent first grade absorption tower amine liquid temperature in, x
6represent two-level absorption tower amine liquid temperature in, x
7represent flash tank pressure, x
8represent reboiler A mouth steam consumption, x
9represent reboiler B mouth steam consumption, x
10represent vapor preheater flow;
Step 2: gather high sulfur-containing natural gas cleaning and desulfurization explained hereafter data, the data obtained is [X
m × N, Y
1, Y
2], wherein: m is input variable number, N is collecting sample quantity, and X is input variable space, Y
1for H
2s content, Y
2for CO
2content;
Gather the technological parameter in high sulfur-containing natural gas cleaning and desulfurization production run, and H2S content in the middle of the rock gas produced under gathering corresponding technological parameter and CO2 content, for follow-up modeling, optimization.
Step 3: pre-service is carried out to the high sulfur-containing natural gas cleaning and desulfurization explained hereafter data of step 2 gained, obtains reflecting the valid data producing actual characteristic;
The sample of default parameters in 3.1 rejecting image data, and ensure that sample meets enterprise's purified gas technical indicator, obtaining new data is [X
m × n, Y
1, Y
2], n is sample size after process, n < N;
3.2 pairs of input variable data carry out the rejecting of gross error data, after gross error data are rejected, and sample is reduced to [X
mH, Y
1, Y
2] (H≤n);
The valid data that can reflect production run actual characteristic can be obtained by the sample and gross error data of rejecting default parameters in image data.
3.3 pairs of input variable data carry out 3 σ criterion process, after 3 σ criterion process, and sample is reduced to [X
mh, Y
1, Y
2] (h≤H);
The basic thought of 3 σ criterion process is: the distance of data upper control limit UCL and lower control limit LCL and center line is the data within 3 σ is usually good.Therefore, the data beyond upper and lower control line are deleted, ensures that data are optimal data.Wherein, the formula of center line and upper and lower control line is as follows:
UCL=μ+3σ,CL=μ,LCL=μ-3σ
Wherein: μ: the mean value of conceptual data; σ: the standard deviation of conceptual data.
To data [X
mH, Y
1, Y
2] each input variable in (H≤n), adopt above-mentioned formula to calculate, determine UCL, CL, LCL.If the value of certain input variable is outside this upper and lower control line, then reject this data sample point, by systematic analysis.
If a large amount of normal value of certain variable is positioned at outside control line, then expands control line scope, to retain the variable of this normal value, obtain new data [X
mh, Y
1, Y
2] (h≤H).
3.4 carry out data normalization process, and obtaining new data is [X'
mh, Y
1', Y
2'];
Adopt method for normalizing, obtain valid data, improve model accuracy.
Step 4: adopt Unscented kalman neural network to pretreated data [X'
mh, Y
1', Y
2'] carry out modeling, to obtain the accurate model of high sulfur-containing natural gas cleaning and desulfurization process production capacity, by Unscented kalman filtering, neural network weight, threshold value are estimated, using neural network weight, threshold value as the state variable of Unscented kalman filtering, the output of neural network as the measurand of Unscented kalman filtering, thus obtains the accurate model of high sulfur-containing natural gas cleaning and desulfurization process production capacity;
Described Unscented kalman neural network is three-layer neural network, and wherein: hidden layer transport function is S type function, output layer transport function is Purelin function, and this three-layer neural network function expression is as follows:
Wherein: M=10, be input layer number; Q is hidden layer neuron number, adopts method of trial and error formula
determine neural network hidden layer neuron number, K is the constant between 0-10, by training pattern effectiveness comparison, selects best q value as neural network hidden layer neuron number;
When adopting Unscented kalman filtering neural network high sulfur-containing natural gas cleaning and desulfurization process productivity model, the initial covariance of state of Unscented kalman filtering, average, and in Unscented kalman filtering in UT conversion the span of spreading factor all given at random in the scope of 0-1;
Unscented kalman filtering algorithm has taken into full account that environmental variance and internal state variable are on the impact of model, can the dynamic tracking characteristics of self-adaptation, propose Unscented kalman Neural Network Online correction static model, realize high sulfur-containing natural gas dynamic evolutionary modeling, improve modeling accuracy.
Step 5: with Unscented kalman filtering neural network model two output variable design preference functions, as fitness function
use multi-objective genetic algorithm, to input variable x
1, x
2, x
3, x
4, x
5, x
6, x
7, x
8, x
9, x
10be optimized in respective top/bottom latitude;
The preference function design of physical layout can realize different physical quantities and design under same measurement criterion, and multi-objective genetic algorithm can provide a series of Pareto optimal solution set of multiple-objection optimization in preference function design basis.
Step 6: bring the h group input variable optimization disaggregation after optimizing into Unscented kalman neural network model successively, calculate model two output valve H now
2s content Y
1, CO
2content Y
2, compare with sample value mean value, peep optimization effect.
Be H in enterprise's purified gas technical indicator purified gas in described step 3.1
2s content is lower than 6mg/m
3, CO
2volume percent content is lower than 3%.
In described step 3.4, concrete normalization processing method is as follows:
Wherein, x
ifor the input variable before normalization, x '
ifor the input variable after normalization, x
minfor input variable x before normalization
iminimum value, x
maxfor the maximal value of input variable before normalization, y
ifor the output variable before normalization, y '
ifor output variable after normalization, y
minfor output variable minimum value before normalization, y
maxfor output variable maximal value before normalization.
Design 3-s class preference function in described step 5, preference is with fabulous, good, generally, poor, extreme difference five descriptive grades, and corresponding numerical intervals is [0,2], [2,4], [4,6], [6,8], [8,10].
Owing to have employed technique scheme, the present invention has following technique effect:
The present invention adopts Kalman filtering neural network to overcome the defect of BP neural network static modelling, make high sulfur-containing natural gas cleaning and desulfurization process can online adaptive modeling, dynamic effects rule between accurate reflection process operation parameter and dbjective state parameter, also realizes the optimization of mutual restricting relation energy consumption and output.
Embodiment
Below in conjunction with drawings and Examples, the invention will be further described:
Embodiment 1:
See Fig. 3, a kind of gas purifying process modeling optimization method based on Unscented kalman neural network, is characterized in that the method is carried out as follows:
Step 1: the input variable determining high sulfur-containing natural gas cleaning and desulfurization model of process: select m the process operation parameter that can be effectively controlled in high sulfur-containing natural gas cleaning and desulfurization art production process as mode input variable, wherein, m=10, input variable is respectively: x
1represent desulfuration absorbing tower amine liquid inlet flow rate, x
2represent tail gas absorber amine liquid inlet flow rate, x
3represent unstripped gas treatment capacity, x
4represent half rich amine solution internal circulating load, x
5represent first grade absorption tower amine liquid temperature in, x
6represent two-level absorption tower amine liquid temperature in, x
7represent flash tank pressure, x
8represent reboiler A mouth steam consumption, x
9represent reboiler B mouth steam consumption, x
10represent vapor preheater flow;
Step 2: gather high sulfur-containing natural gas cleaning and desulfurization explained hereafter data, the data obtained is [X
m × N, Y
1, Y
2], wherein: m is input variable number, N is collecting sample quantity, and X is input variable space, Y
1for H
2s content, Y
2for CO
2content;
Gather the technological parameter in high sulfur-containing natural gas cleaning and desulfurization production run, and H2S content in the middle of the rock gas produced under gathering corresponding technological parameter and CO2 content, for follow-up modeling, optimization.
Step 3: pre-service is carried out to the high sulfur-containing natural gas cleaning and desulfurization explained hereafter data of step 2 gained, obtains reflecting the valid data producing actual characteristic;
The sample of default parameters in 3.1 rejecting image data, and ensure that sample meets enterprise's purified gas technical indicator, obtaining new data is [X
m × n, Y
1, Y
2], n is sample size after process, n < N;
Be H in enterprise's purified gas technical indicator purified gas in described step 3.1
2s content is lower than 6mg/m
3(Y
1<4), CO
2volume percent content is lower than 3% (Y
2<3).
3.2 pairs of input variable data carry out the rejecting of gross error data, after gross error data are rejected, and sample is reduced to [X
mH, Y
1, Y
2] (H≤n);
The valid data that can reflect production run actual characteristic can be obtained by the sample and gross error data of rejecting default parameters in image data.
3.3 pairs of input variable data carry out 3 σ criterion process, after 3 σ criterion process, and sample is reduced to [X
mh, Y
1, Y
2] (h≤H);
The basic thought of 3 σ criterion process is: the distance of data upper control limit UCL and lower control limit LCL and center line is the data within 3 σ is usually good.Therefore, the data beyond upper and lower control line are deleted, ensures that data are optimal data.Wherein, the formula of center line and upper and lower control line is as follows:
UCL=μ+3σ,CL=μ,LCL=μ-3σ
Wherein: μ: the mean value of conceptual data; σ: the standard deviation of conceptual data.
To data [X
mH, Y
1, Y
2] each input variable in (H≤n), adopt above-mentioned formula to calculate, determine UCL, CL, LCL.If the value of certain input variable is outside this upper and lower control line, then reject this data sample point, by systematic analysis.
If a large amount of normal value of certain variable is positioned at outside control line, then expands control line scope, to retain the variable of this normal value, obtain new data [X
mh, Y
1, Y
2] (h≤H).
3.4 carry out data normalization process, and obtaining new data is [X'
mh, Y
1', Y
2'];
In described step 3.4, concrete normalization processing method is as follows:
Wherein, x
ifor the input variable before normalization, x '
ifor the input variable after normalization, x
minfor input variable x before normalization
iminimum value, x
maxfor the maximal value of input variable before normalization, y
ifor the output variable before normalization, y '
ifor output variable after normalization, y
minfor output variable minimum value before normalization, y
maxfor output variable maximal value before normalization.
Adopt method for normalizing, obtain valid data, improve model accuracy.
Step 4: adopt Unscented kalman neural network to pretreated data [X'
mh, Y
1', Y
2'] carry out modeling, to obtain the accurate model of high sulfur-containing natural gas cleaning and desulfurization process production capacity, by Unscented kalman filtering, neural network weight, threshold value are estimated, using neural network weight, threshold value as the state variable of Unscented kalman filtering, the output of neural network as the measurand of Unscented kalman filtering, thus obtains the accurate model of high sulfur-containing natural gas cleaning and desulfurization process production capacity;
Described Unscented kalman neural network is three-layer neural network, and wherein: hidden layer transport function is S type function, output layer transport function is Purelin function, and this three-layer neural network function expression is as follows:
Wherein: M=10, be input layer number; Q is hidden layer neuron number, adopts method of trial and error formula
determine neural network hidden layer neuron number, K is the constant between 0-10, by training pattern effectiveness comparison, selects best q value as neural network hidden layer neuron number;
When adopting Unscented kalman filtering neural network high sulfur-containing natural gas cleaning and desulfurization process productivity model, the initial covariance of state of Unscented kalman filtering, average, and in Unscented kalman filtering in UT conversion the span of spreading factor all given at random in the scope of 0-1;
Unscented kalman filtering algorithm has taken into full account that environmental variance and internal state variable are on the impact of model, can the dynamic tracking characteristics of self-adaptation, propose Unscented kalman Neural Network Online correction static model, realize high sulfur-containing natural gas dynamic evolutionary modeling, improve modeling accuracy.
Step 5: with Unscented kalman filtering neural network model two output variable design preference functions, as fitness function
use multi-objective genetic algorithm, to input variable x
1, x
2, x
3, x
4, x
5, x
6, x
7, x
8, x
9, x
10be optimized in respective top/bottom latitude;
Design 3-s class preference function in described step 5, preference is with fabulous, good, generally, poor, extreme difference five descriptive grades, and corresponding numerical intervals is [0,2], [2,4], [4,6], [6,8], [8,10].
The preference function design of physical layout can realize different physical quantities and design under same measurement criterion, and multi-objective genetic algorithm can provide a series of Pareto optimal solution set of multiple-objection optimization in preference function design basis.
Step 6: bring the h group input variable optimization disaggregation after optimizing into Unscented kalman neural network model successively, calculate model two output valve H now
2s content Y
1, CO
2content Y
2, compare with sample value mean value, peep optimization effect.
Analyze with the production data of certain high sulfur-containing natural gas purification plant desulfurizer, monitor data is from DCS system.
Step 1-2: choose 744 groups of data of some moons, reject by the correction of supervisory system and data corresponding to observational error, ensure that sample meets enterprise's purified gas technical indicator, namely H2S content is lower than 4ppmv, CO
2content is lower than 3%.Sampled data is as shown in table 1.
Certain high sulfur-containing natural gas purification plant desulfurizer data list of table 1
Step 3: his-and-hers watches 1 data carry out pre-service, obtain data [X'
mh, Y
1', Y
2'] as shown in table 2.
Data after table 2 normalized
Step 4: adopt Unscented kalman neural network high sulfur-containing natural gas cleaning and desulfurization model of process, by desulfuration absorbing tower amine liquid inlet tube flow x in high sulfur-bearing sulfur removal technology
1(t/h), tail gas absorber amine liquid inlet tube flow x
2(t/h), unstripped gas treatment capacity x
3(kNm3/h), the internal circulating load x of half rich amine solution
4(t/h), first grade absorption tower amine liquid temperature in x
5(DEG C), two-level absorption tower amine liquid temperature in x
6(DEG C), flash tank pressure x
7(MPa), reboiler A inlet vapor consumption x
8(kg/h), reboiler B inlet vapor consumption x
9and vapor preheater flow x (kg/h)
10(t/h) these 10 operating parameters are as the input of statistical modeling, by desulfurization unit product H
2s content (ppmv) and CO
2content (%) exports, to data [X' as target
mh, Y
1', Y
2'] analyze, estimated neural network weight, threshold value by Unscented kalman filtering, using neural network weight, threshold value as the state variable of Unscented kalman filtering, the output of neural network is as the measurand of Unscented kalman filtering.Thus obtain the accurate model of high sulfur-containing natural gas cleaning and desulfurization technological process operating conditions and purified gas acidic components.
When adopting Unscented kalman filtering neural network high sulfur-containing natural gas cleaning and desulfurization model of process, only need to provide the initial covariance of state, the average of Unscented kalman filtering first, and in Unscented kalman filtering UT conversion in spreading factor, given at random in the scope of 0-1.Wherein the parameter of UT conversion is: k=0; α=0.06; β=0.4.
Input in order by above-mentioned sample, select 724 groups of data as training sample, 362 groups of data are as its test samples.The Unscented kalman neural net model establishing training sample obtained, training and testing tracking power and error effects respectively as Fig. 4,5,6, shown in 7.
Can find out, use Unscented kalman neural network to carry out online adaptive dynamic modeling to high sulfur-containing natural gas cleaning and desulfurization process and achieve good effect.Model accuracy and the ideal value of Unscented kalman neural network overlap substantially, establish high-precision high sulfur-containing natural gas cleaning and desulfurization process dynamic model.Unscented kalman neural network model precision is more satisfactory.
The high sulfur-containing natural gas cleaning and desulfurization process of Unscented kalman neural network to complexity is adopted to carry out online adaptive dynamic modeling, its tracking accuracy obtains effective guarantee, effectively establishes the high sulfur-containing natural gas cleaning and desulfurization process online adaptive model approached based on Unscented kalman neural network subspace.
Step 5:H
2s and CO
2taken in excess energy consumption can be caused greatly to increase.In conjunction with numerical analysis, and consider the actual condition of sulfur removal technology, intend at original H
2s content and CO
2on content basis, there is suitable raising, get and work as H
2s content and CO
2content is respectively 3ppmv, and 2.2%, sulfur removal technology reaches optimum efficiency.
Design 3-s class preference function, as shown in Figure 1.Preference is with fabulous, good, generally, poor, extreme difference five descriptive grades, and corresponding numerical intervals is [0,2], [2,4], [4,6], [6,8], [8,10].
H
2s content preference function is
CO
2content preference function is
Step 6: on Unscented kalman neural network accurate high sulfur-containing natural gas cleaning and desulfurization process modeling basis, carry out MOGA multiple-objection optimization.Wherein, the internal circulating load of poor amine liquid, unstripped gas treatment capacity, the internal circulating load of half rich amine solution, first grade absorption tower amine liquid temperature in, two-level absorption tower amine liquid temperature in, flash tank pressure, reboiler steam consumption, reboiler steam consumption, the optimization range of vapor preheater flow 10 input variables is respectively: [229.975, 270.761], [184.140, 204.601], [98.720, 130.008], [239.287, 272.457], [32.168, 36.369], [30.728, 38.030], [0.647, 0.651], [17.345, 30.663], [22.145, 28.119], [0, 1.431].Arranging optimization population scale is 30, genetic algebra 50, and the Pareto obtained optimizes forward position and separates, as shown in table 3.
Table 3 MOGA process parameter optimizing result
For high sulfur-containing natural gas cleaning and desulfurization technological process, as shown in table 4 through optimizing acidic components, output and energy consumption in after purification gas.
The performance index of table 4 prioritization scheme
Prioritization scheme can realize purified gas H2S content and bring up to 1.04ppmv from 0.3ppmv, and CO2 content brings up to 1.64% from original 1.47%.The taken in excess of sulfur removal technology to sour gas makes moderate progress.For the natural gas conditioning industry of day output 3,000 ten thousand side, enterprise's day output will improve (107.87-98.95) × 24=214.08KNm3
What finally illustrate is, above preferred embodiment is only in order to illustrate technical scheme of the present invention and unrestricted, although by above preferred embodiment to invention has been detailed description, but those skilled in the art are to be understood that, various change can be made to it in the form and details, and not depart from claims of the present invention limited range.