CN104656441A - Natural gas purification process modeling optimization method based on unscented kalman neural network - Google Patents

Abstract

The invention aims to overcome the disadvantages in the prior art, and provides a natural gas purification process modeling optimization method based on an unscented kalman neural network. The natural gas purification process modeling optimization method comprises the following steps: determining input variables; collecting process production data; carrying out preprocessing on the data; carrying out data normalization processing; adopting the unscented kalman neural network to carry out modeling on the data to obtain a model; designing a preference function by using two output variables of the model of the unscented kalman neural network, and applying a multi-target genetic algorithm to optimize the input variables; disaggregating the input variables after being optimized and bringing in the model of the unscented kalman neural network in sequence, calculating the two output values of the model at the time, comparing with a sample value average value, and observing the optimization effect. The method can establish an accurate and reliable high sulfur natural gas purification desulfurization industrial process model, the yield of the finished product can be improved based on the model, the energy consumption in the desulfurization process can be reduced, and the method has important practical significance for guiding practical industrial production.

Description

Based on the gas purifying process modeling optimization method of Unscented kalman neural network

Technical field

The invention belongs to intelligent power saving yield-increasing technology in high sulfur-containing natural gas desulfurization production run, relate to a kind of gas purifying process modeling optimization method based on Unscented kalman neural network.

Background technology

High sulfur-containing natural gas industrial flow is complicated, and process parameter is numerous, and affecting by uncertain factors such as temperature, pressure, flow, ageing equipment and unstripped gas treatment capacities, is typical complex nonlinear dynamic perfromance chemical system.High sulfur-containing natural gas cleaning and desulfurization process mainly comprises with lower part: main absorption tower MDEA solution absorbs acidic components H2S and CO2, hydrolysis reactor removes (COS), the cyclic regeneration of regenerating column MDEA solution and heat transfer process, concrete technology flows through journey as shown in Figure 2.How setting up accurately reliable high sulfur-containing natural gas cleaning and desulfurization industrial process model is improve finished product gas output, reducing basis and the prerequisite of sweetening process energy consumption, having important practical significance to instructing actual industrial production.

The mechanism model of high sulfur-containing natural gas cleaning and desulfurization process can describe the variation tendency of significant variable in production, the mechanism knowledge of reflection production run.But high sulfur-containing natural gas cleaning and desulfurization production run is complicated physics, a chemical process, generally have complex structure, multivariate, non-linear, the feature such as time lag, uncertainty, traditional modelling by mechanism method is difficult to the requirement meeting Accurate Model.Neural network (Artificial Neural Network, ANN) with its powerful None-linear approximation ability, compared with traditional mechanisms modeling method, belong to statistical modeling method, have the advantages that can set up and not rely on accurate process principle He can approach any Nonlinear Mapping with arbitrary accuracy.

ANN demonstrates unique superiority on the modeling problem of process complication system, is widely used in industrial process modeling.But at present, during application neural network high sulfur-containing natural gas cleaning and desulfurization production run model, have ignored environmental variance and internal state variable to the impact of model, suppose that its neighbourhood noise and internal state variable are metastable.Often just carrying out simple static mappings to input/output variable, is a kind of static state modeling method, to high sulfur-containing natural gas cleaning and desulfurization production run modeling limited efficiency.The high-precision model how setting up high sulfur-containing natural gas cleaning and desulfurization process becomes difficult point.

UKF neural network adopts to be had self-adaptation dynamic tracking capabilities UKF filtering algorithm and adjusts static neural network model, and using neural network weight and the threshold value state variable as UKF, the output of neural network is as the measurand of UFN.Set up the model of dynamic realtime filter effect by state estimation, the actual change situation of system can be reflected, to obtain accurate model.

In natural gas conditioning desulfurization production run, energy consumption and output are two important performance assessment criteria.But there is again mutual restricting relation between output and energy consumption, and must to sacrifice another target as cost to one of them objective optimization, and the unit of each target is often inconsistent, is therefore difficult to the superiority-inferiority evaluating two target problem solutions objectively.

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:

$y=h(w_k,x_k)=F^2(w_k^2,F^1(w_k^1,x_k))=\underset{i=1}{\overset{q}{\mathrm{\&Sigma;}}}\frac{w_i^2}{1+e[\underset{j=1}{\overset{M}{\mathrm{\&Sigma;}}}{w}_{\mathrm{ij}}{x}_i+b_{1i}]}+b_q$

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:

$x_i^{\&prime;}=\frac{x_i-x_{\min }}{x_{\max }-x_{\min }},y_i^{\&prime;}=\frac{y_i-y_{\min }}{y_{\max }-y_{\min }},$

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.

Accompanying drawing explanation

Fig. 1 is that sulfur removal technology optimizes multiple goal preference function curve map;

Fig. 2 is certain high sulfur-containing natural gas desulfurization simulation process process flow diagram;

Fig. 3 is process flow diagram of the present invention;

Fig. 4 is UKFNN neural network model H2S training effect figure;

Fig. 5 is that UKFNN neural network model H2S tests design sketch;

Fig. 6 is UKFNN neural network model CO2 training effect figure;

Fig. 7 is that UKFNN neural network model CO2 tests design sketch.

Reference numeral

In Fig. 2,1 hydrolysis reactor fed separator; 2 hydrolysis reactor primary heaters; 3 hydrolysis reactors; 4 hydrolysis reactor entry/exit material heat interchanger; 5 hydrolysis reactor aftercoolers; 6 two-level absorption towers; 7 first grade absorption towers; 8 poor amine liquid pumps; 9 middle amine liquid pumps; 10 middle amine liquid refrigeratorys; 11 poor amine liquid aftercoolers; 12 regenerating columns; 13 regeneration overhead air coolers; 14 amine liquid regeneration overhead return tanks; 15 sour water reflux pumps; Poor amine liquid pump at the bottom of 16 regenerating columns; 17 poor rich liquid heat exchangers; 18 poor amine liquid air coolers; 19 amine liquid flash tanks; Feed gas: raw natural gas; Treated gas: purified gas; Acid gas: acid gas.

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:

$x_i^{\&prime;}=\frac{x_i-x_{\min }}{x_{\max }-x_{\min }},y_i^{\&prime;}=\frac{y_i-y_{\min }}{y_{\max }-y_{\min }},$

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:

$y=h(w_k,x_k)=F^2(w_k^2,F^1(w_k^1,x_k))=\underset{i=1}{\overset{q}{\mathrm{\&Sigma;}}}\frac{w_i^2}{1+e[\underset{j=1}{\overset{M}{\mathrm{\&Sigma;}}}{w}_{\mathrm{ij}}{x}_i+b_{1i}]}+b_q$

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 $\mathrm{perf}\left(y_1\right)=\left\{\begin{array}{c}1.1111*{(y_1-3)}^2,0<y_1\&le;3\\ 10*{(y_1-3)}^2,3<y_1\&le;4\end{array}\right.$

CO ₂content preference function is $\mathrm{perf}\left(y_2\right)=\left\{\begin{array}{c}20661*{(y_1-2.2\%)}^2,0<y_2\&le;2.2\%\\ 156250*{(y_1-2.2\%)}^2,2.2\%<y_2\&le;3\%\end{array}\right.$

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.

Claims (4)

1., 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;

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);

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);

3.4 carry out data normalization process, and obtaining new data is [X' _mh, Y ₁', Y ₂'];

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;

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;

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.

2. the gas purifying process modeling optimization method based on Unscented kalman neural network according to claim 1, is characterized in that: 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%.

3. the gas purifying process modeling optimization method based on Unscented kalman neural network according to claim 1, is characterized in that: 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.

4. the gas purifying process modeling optimization method based on Unscented kalman neural network according to claim 1, is characterized in that: design 3-s class preference function in described step 5, preference is with fabulous, good, generally, poor, extreme difference five descriptive grades, corresponding numerical intervals is [0,2], [2,4], [4,6], [6,8], [8,10].