CN106777866A - Technology Modeling and optimization method are purified towards energy-saving high sulfur-containing natural gas - Google Patents

Abstract

What the present invention was provided purifies Technology Modeling and optimization method towards energy-saving high sulfur-containing natural gas, including：It is acquired to form sample set after choosing the technological parameter of influence desulfuration efficiency and the performance indications of desulfurization unit；Sample set is normalized to form normalization sample set, and therefrom chooses training sample and test sample；Neural network model is built based on training sample and the original state variable of neural network model is determined；The optimum state variable of neural network model is estimated with ST UKFNN algorithms；Neural network model is updated according to optimum state variable；H is built respectively₂S concentration and CO₂The preference function of concentration；Using MOGA algorithms to H₂S concentration and CO₂The technological parameter of concentration is optimized, and the neural network model that the technological parameter after optimization is brought into after updating, the systematic function of the technological parameter after calculation optimization, the average value with the systematic function of actual sample is compared.The production efficiency of high sulfur-containing natural gas purification can be improved using the present invention.

Description

Technology Modeling and optimization method are purified towards energy-saving high sulfur-containing natural gas

Technical field

The present invention relates to high sulfur-containing natural gas purification techniques field, more specifically, it is related to a kind of towards energy-saving High sulfur-containing natural gas purify Technology Modeling and optimization method.

Background technology

High sulfur-containing natural gas acidic components content is higher by several times than conventional gas, its sweetening process amine liquid internal circulating load is big, Technological process is complicated, high energy consumption.Statistics shows that desulfurization unit energy consumption accounts for high sulfur-containing natural gas purification plant total energy consumption more than 50%, Its unit comprehensive energy consumption is up to 1729.3MJt-1, belongs to highly energy-consuming unit.For large-scale purification plant, by desulfurization unit Optimize capable of reducing energy consumption 5%~10%.Additionally, high sulfur-containing natural gas acidic components concentration is high, by the product tolerance after purification Relative raw material throughput is decreased significantly.Therefore, carrying out process optimization to high sulfur-containing natural gas sweetening process, realize that energy-conservation drops Consumption, it is very necessary to improve yield and gas processing economic benefit.

The content of the invention

In view of the above problems, it is an object of the invention to provide a kind of the dynamic of high sulfur-containing natural gas purification based on ST-UKFNN State evolutionary Modeling optimization method, to solve the problems, such as that above-mentioned background technology is proposed.

What the present invention was provided purifies Technology Modeling and optimization method towards energy-saving high sulfur-containing natural gas, including：

Step S1：The technological parameter of selection influence desulfuration efficiency and the performance indications of desulfurization unit；Wherein, technological parameter bag Include into the poor amine flow quantity x of tail gas absorber₁, into the poor amine flow quantity x of two-level absorption tower₂, unstripped gas treating capacity x₃, tail Gas unit returns to half rich amine flow quantity x of desulfurization unit₄, first grade absorption tower amine liquid enter tower temperature degree x₅, two-level absorption tower amine liquid enter tower Temperature x₆, flash tank pressure x₇, reboiler steam consumption x₈, another reboiler steam consumption x₉And steam The steam consumption x of preheater₁₀；The performance indications of desulfurization unit include H in purified gas₂The concentration of SAnd CO₂ConcentrationAnd the yield y of purified gas_pg；

Step S2：The technological parameter of Preset Time and the data of performance indications are gathered, sample is formed after rejecting error sample Collection [X, Y]；

Step S3：Sample set [X, Y] is normalized, normalization sample set is formedTake normalization sample setIn preceding 80% sample as training sample, and remaining 20% sample is used as test sample；

Step S4：The original state variable X of neural network model and neural network model is built based on training sample, with And, by training sampleAs the input of neural network model, by training sampleAs neural network model Output；

Wherein, the neural network model of structure is：

Wherein, I_kIt is the vector sample value of training sample, and as the input of neural network model,It is network input layer To the connection weight of the neuron of hidden layer,It is the threshold value of the neuron of network input layer to hidden layer,For hidden layer is arrived The connection weight of the neuron of network output layer,It is hidden layer to the threshold value of the neuron of network output layer, wherein, i=1, 2…S₀；J=1,2 ... S₁；K=1,2 ... S₂；S₀It is the quantity of the neuron of network input layer, S₁It is the neuron of network hidden layer Quantity, S₂It is the quantity of the neuron of network output layer；

The original state variable of structure is：

Step S5：The optimum state variable of neural network model is estimated using ST-UKFNN algorithms；

Step S6：Using optimum state variable as the neural network modelWithTo formula (1) It is updated, obtains the neural network model after training sample updates；

Step S7：H is built respectively₂S concentrationPreference function and CO₂ConcentrationPreference function；

Step S8：Using MOGA algorithms respectively to H₂S concentrationPreference function and CO₂ConcentrationPreference function enter The optimizing of row multiple target extreme value optimizes, and acquisition meets the decision variable of produce reality；

Step S9：The neural network model that decision variable after optimization is brought into after training sample updates, after calculation optimization Decision variable systematic function, the average value with the systematic function of actual sample is compared, if the decision-making after optimization becomes The systematic function of amount is carried out using the decision variable after optimization more than the average value of the systematic function of actual sample to actual production Instruct；Otherwise repeat the above steps S1-S8, until the systematicness of the systematic function more than actual sample of the decision variable after optimization Untill the average value of energy.

What the present invention was provided purifies Technology Modeling and optimization method towards energy-saving high sulfur-containing natural gas, can be effective The whole optimal economic benefit route of ground tracks of device, effectively overcomes process interference, equipment performance change, economic benefit and production The variation issue of target, thus it is energy-saving, improve yield and gas processing economic benefit.

Brief description of the drawings

By reference to the explanation below in conjunction with accompanying drawing and the content of claims, and with to it is of the invention more comprehensively Understand, other purposes of the invention and result will be more apparent and should be readily appreciated that.In the accompanying drawings：

Fig. 1 is H in purified gas₂The preference function figure of S concentration；

Fig. 2 is CO in purified gas₂The preference function figure of concentration；

Fig. 3 is H₂The content preference function and CO of S concentration₂The Pareto disaggregation figures of the content preference function of concentration；

Fig. 4 is H₂The content actual function and CO of S concentration₂The Pareto disaggregation figures of the content actual function of concentration.

Specific embodiment

Explanation of nouns

ST-UKFNN：Strong track Unscented Kalman Fliter Neural Network, follow the trail of by force nothing Mark Kalman filtering neutral net.

What the present invention was provided purifies Technology Modeling and optimization method towards energy-saving high sulfur-containing natural gas, including：

Step S1：The technological parameter of selection influence desulfuration efficiency and the performance indications of desulfurization unit；Wherein, technological parameter bag Include into tail gas absorber poor amine flow quantity, poor amine flow quantity, unstripped gas treating capacity, tail gas unit into two-level absorption tower Half rich amine flow quantity of return desulfurization unit, first grade absorption tower amine liquid enter tower temperature degree, two-level absorption tower amine liquid and enter tower temperature degree, flash distillation The steam consumption of pressure tank, the steam consumption, the steam consumption of another reboiler and vapor preheater of reboiler Amount；The performance indications of desulfurization unit include H in purified gas₂S and CO₂Concentration and purified gas yield.As shown in table 1：

Table 1

Step S2：The technological parameter of Preset Time and the data of performance indications are gathered, sample is formed after rejecting error sample Collection [X, Y].Sample set [X, Y] is as shown in table 2 below：

Table 2

Step S3：Sample set [X, Y] is normalized, normalization sample set is formedTake normalization sample setIn preceding 80% sample as training sample, and remaining 20% sample is used as test sample.

Step S4：The original state variable X of neural network model and neural network model is built based on training sample, with And, by training sampleInput as neural network model isBy in training sampleAs nerve net The output of network model is

Wherein, the neural network model of structure is：

Wherein, I_kIt is the vector sample value of training sample, and as the input of neural network model,It is network input layer To the connection weight of the neuron of hidden layer,It is the threshold value of the neuron of network input layer to hidden layer,For hidden layer is arrived The connection weight of the neuron of network output layer,It is hidden layer to the threshold value of the neuron of network output layer, wherein, i=1, 2…S₀；J=1,2 ... S₁；K=1,2 ... S₂；S₀It is the quantity of the neuron of network input layer, S₁It is the neuron of network hidden layer Quantity, S₂It is the quantity of the neuron of network output layer；

The original state variable of structure is：

Step S5：The optimum state variable of neural network model is estimated using ST-UKFNN algorithms.

The present invention estimates the state variable of neural network model using ST-UKFNN algorithms, to reach connection weight, threshold value Continuous adjustment, until meet require.The state estimation of the optimum state variable that will be obtained is used as above-mentioned set up neutral net The connection weight of model, threshold value.It should be noted that the connection weight, threshold value are the company after ST-UKFNN algorithms are adjusted Weights, threshold value are connect, is also whole connection weights and threshold value of above-mentioned set up neural network model, including With

The process of the optimum state variable of neural network model is estimated using ST-UKFNN algorithms to be included：

Step S51：Sigma samplings are carried out to original state variable X, 2n+1 sampled point, initialization control 2n+1 is obtained The distribution parameter alpha of individual sampled point, parameter κ to be selected, and non-negative right factor beta, the Sigma samplings to original state variable X It is as follows：

Wherein,It is the i-th row of the optimum state variable at (k-1) moment, n is state matrix dimension, p_k-1It is (k- 1) covariance of the optimum state variable at moment.

Step S52：The weight of each sampled point is calculated, the weight of each sampled point is as follows：

Wherein, W_cTo calculate the weight of the covariance of state variable, W_mPower during to calculate state estimation and observation prediction Weight,It isFirst row,It isFirst row.

Step S53：It is optimal by (k-1) moment of each sampled point by the state equation of Discrete time Nonlinear Systems The state estimation of state variable is transformed to the state estimation of the state variable at k momentAnd, by merging the k moment State estimationVector, obtain the k moment state variable state prior estimateAnd covariance

The state estimation of the state variable at k momentFor：

Wherein, w_kIt is process noise, its covariance matrix Q_kIt is cov (w_k,w_j)=Q_kδ_kj,

The state prior estimate of the state variable at k momentFor：

The covariance P of the state variable at k moment_k|k-1For：

Step S54：The state of the state variable that will set up the k moment by the observational equation of Discrete time Nonlinear Systems is estimated MeterObservation with the k moment is predictedContact with complete observation prediction, and estimate the k moment observation prediction association side Difference P_yk。

The average of the observation prediction at k momentFor：

Wherein,

Above-mentioned formula (8) and formula (9) establish the state estimation of the state variable at k momentObservation with the k moment is predicted EstimateBetween relation.

Wherein, ν_kIt is observation noise, its covariance matrix R_kIt is cov (v_k,v_j)=R_kδ_kj,

The covariance of the observation prediction at k momentFor：

Wherein, strong tracing algorithm, i.e. fading factor λ are introduced herein_k+1Strengthen the trace ability of model to improve model essence Degree；

N_k+1=V_k+1-βR_k+1 (14)

Wherein, β is the reduction factor, β >=1；

Step S55：Calculate the covariance P between the state variable at k moment and observation prediction_xy,k：

Step S56：By setting up covariance P_xy,kAnd covarianceRelation, update the k moment state variable state Estimate and covariance, obtain the optimum state variable at k moment.

Wherein, the covariance P of the state variable of foundation_xy,kWith the covariance of observation predictionRelation be：

Wherein, K_kIt is gain matrix, when realizing the state estimation of the optimum state variable for updating the k moment and update k with this The covariance P of the state variable at quarter_k；And,

The state estimation X of the state variable at the k moment after renewal_k|kFor：

The covariance P of the state variable at the k moment after renewal_kFor：

By the state estimation X of the state variable at the k moment after renewal_kWith covariance P_kAs the optimal variable at k moment.

The optimum state variable at the k moment of acquisition is substituted into step S51 and re-starts sigma samplings, circulation step by step S57 Rapid S51-S57, obtains the optimum state variable of neural network model.

The structural parameters of the optimum state variable of neural network model are as follows：

(1)H₂S concentration model structure parameters

w₂(1 × 20)=[- 0.42 0.22 ... 0.14-0.49] b₂(1 × 1)=[- 0.19]

(2)CO₂Concentration model structure parameter

w₂(1 × 20)=[- 0.46 4.74 ... 0.15 0.27] b₂(1 × 1)=[- 0.10]

Step S6：Using optimum state variable as neural network modelWithFormula (1) is carried out Update, obtain the neural network model after training sample updates.

Step S7：H is built respectively₂S concentrationAnd CO₂ConcentrationPreference function.

In system process parameters optimization is calculated, it is considered to which designer has different fancy grades to different parameters, utilizes Physical layout constructing system preference function.Setting Liquid output optimal value y_1best, setting value y_bestThe a certain contiguous range of surrounding [y_best-△y,y_best+ △ y] in fluctuation be very satisfied (HD), and in [y_best-△y-△y₁,y_best-△y],[y_best+△y, y_best+△y+△y₁] in be satisfied (D), (T) is subjected to successively, be unsatisfied with it is (U) and very dissatisfied (HU), it is corresponding inclined Good value is interval to use [0,2], and [2,4], [4,6], [6,8], [8,10] represent.

Build the functional relation perf of preference value and actual amount_c(H₂) and perf S_c(CO₂), and with preference numerical value area Between corresponding H₂S concentration and CO₂Concentration ranges value is as shown in table 3 below：

The preference numerical intervals of table 3 interval table corresponding with variable actual value

The corresponding relation of the interval border value according to table 3, can be in the hope of H₂S concentration, CO₂The preference function expression formula of concentration As shown in formula 21, formula 22：

H₂The preference function of S concentration is as shown in figure 1, CO₂The preference function of concentration is as shown in Figure 2.

Step S8：Using MOGA algorithms respectively to H₂S concentrationPreference functionAnd CO₂Concentration y_CO2It is inclined Good functionMultiple target extreme value optimizing optimization is carried out, acquisition meets the decision variable of produce reality.

The specific optimization process of step S8, including：

Step S81：By the technological parameter x before optimization₁-x₁₀Respectively as decision variable, by decision variable P=[x₁ x₂ L x₁₀] fitness function value comparing find optimized individual；Wherein, the performance variable function of part maximizing is carried out instead Normalization obtains fitness function：

The performance variable function of part maximizing：

Due to the H for during optimization, obtaining₂S concentration, CO₂Concentration is better closer to optimum value, so by formula (24) Carry out renormalization and obtain formula (23).

Step S82：Using decision variable P=[x₁ x₂ L x₁₀] parent population P is built, wherein,

Wherein, K is the individuality in parent population PQuantity；L is the population sample of initialization Quantity, L=50；GEN is maximum genetic algebra, GEN=100；

Step S83：Bound x according to decision variable_i,min≤x_i≤x_i,max(i=1,2, L, 10) initialization father population P； Wherein, the process of initialization father population P is：From the poor amine flow quantity x of entrance tail gas absorber₁Span in random value AssignFrom the poor amine flow quantity x for entering two-level absorption tower₂Span in random value assign Until the steam consumption x from vapor preheater₁₀Span in random value assign

Step S84：Father population P to initializing carries out first time genetic iteration (GEN=1) to produce population of future generation.

Father population P to initializing carries out the process of first time genetic iteration, including：

Step S841：According to the being dominant property that each is solved in the father population P that fitness function inspection is initialized；Wherein, for One solution i, its grade r_iEqual to 1 plus the number n better than solution i_i, i.e. r_i=n_i+1。

Due to there is no the solution better than noninferior solution in the father population P of initialization, so the grade of noninferior solution is equal to 1.

Step S842：In the population P that will be initialized it is all it is individual be layered according to grade ascending order, then by with one it is linear (or other) respective function to each individuality distribution one initial adaptive value.

Generally select make distribution the adaptive value N individuality of grade (correspondence optimal) and 1 (the correspondingly individuality of worst grade) it Between respective function.

Step S843：The average value of the initial adaptive value of each individuality each grade Nei is calculated, the average value is each etc. The specified adaptive value of each individuality in level.

Step S844：By formula (26) calculate standardization in any one grade between any two individuality i and j away from From：

Wherein,WithIt is k-th maximum and minimum value of object function.

Step S845：Calculated by formula (27) has same grade r with solution i_iEach solution d_ij；

Wherein, α=1 is Sharing Function, σ_shareIt is default microhabitat radius；

Each individual microhabitat number is the summation of Sharing Function value in the grade：

Wherein, μ (r_i) it is that all grades are r_iNumber of individuals.

For the diversity for keeping being solved in noninferior solution, microhabitat number is introduced in the individuality of each grade.

Step S846：The specified adaptive value of each individuality is obtained into the shared suitable of each individuality divided by respective microhabitat number Should be worth.

Step S847：Change of scale is done to all individual shared adaptive values in each grade.

It is to keep the average of all individualities of each grade to the purpose that individual shared adaptive value does dimensional variation Shared adaptive value is identical with average specified adaptive value, i.e., each individual identical probability is chosen to use.

Step S848：Ratio selection, single-point is carried out to each grade by change of scale to intersect, make a variation under calculating acquisition Generation population.

Step S85：Gen=gen+1, circulates 100 step S83~step S84, obtains GEN for population as optimization Result is exported,

Obtain Pareto disaggregation, H₂The Pareto disaggregation of the content preference function of S concentration is as shown in figure 3, CO₂Concentration contains The Pareto disaggregation for measuring preference function is as shown in Figure 4.

Contrast on effect is as shown in table 4 before and after understanding optimization by optimization gained Pareto solution set analysis：

Table 4

Optimum results are the technological parameter x after optimization₁-x₁₀。

Technological parameter x after optimization₁-x₁₀Bound it is as shown in table 5：

Table 5

Step S9：The neural network model that decision variable after optimization is brought into after training sample updates, after calculation optimization Decision variable systematic function, the average value with the systematic function of actual sample is compared, if the decision-making after optimization becomes The systematic function of amount is carried out using the decision variable after optimization more than the average value of the systematic function of actual sample to actual production Instruct；Otherwise repeat the above steps S1-S8, until the systematicness of the systematic function more than actual sample of the decision variable after optimization Untill the average value of energy.

The actual technological parameter x for after optimization of decision variable after optimization₁-x₁₀。

The above, specific embodiment only of the invention, but protection scope of the present invention is not limited thereto, and it is any Those familiar with the art the invention discloses technical scope in, change or replacement can be readily occurred in, should all contain Cover within protection scope of the present invention.Therefore, protection scope of the present invention described should be defined by scope of the claims.

Claims (4)

1. it is a kind of to purify Technology Modeling and optimization method towards energy-saving high sulfur-containing natural gas, including：

Step S1：The technological parameter of selection influence desulfuration efficiency and the performance indications of desulfurization unit；Wherein, the technological parameter bag Include into the poor amine flow quantity x of tail gas absorber₁, into the poor amine flow quantity x of two-level absorption tower₂, unstripped gas treating capacity x₃, tail Gas unit returns to half rich amine flow quantity x of desulfurization unit₄, first grade absorption tower amine liquid enter tower temperature degree x₅, two-level absorption tower amine liquid enter tower Temperature x₆, flash tank pressure x₇, reboiler steam consumption x₈, another reboiler steam consumption x₉And steam The steam consumption x of preheater₁₀；The performance indications of the desulfurization unit include H in purified gas₂The concentration of SAnd CO₂It is dense DegreeAnd the yield y of purified gas_pg；

Step S2：The technological parameter of Preset Time and the data of the performance indications are gathered, is formed after rejecting error sample Sample set [X, Y]；

Step S3：Sample set [X, Y] is normalized, normalization sample set is formedTake the normalization sample setIn preceding 80% sample as training sample, and remaining 20% sample is used as test sample；

Step S4：The original state variable X of neural network model and the neural network model is built based on the training sample, And, by the training sampleAs the input of the neural network model, by the training sampleAs The output of the neural network model；

Wherein, the neural network model is：

$Y=\underset{j=1}{\overset{s_2}{\mathrm{\Σ}}}\left(f\left(\underset{i=1}{\overset{S_1}{\mathrm{\Σ}}}{w}_{ik}^1{I}_k+b_i^1\right)\right)\·w_{kj}^2+b_j^2---\left(1\right)$

Wherein, I_kIt is the vector sample value of the training sample, and as the input of the neural network model,For network is defeated Enter layer to the connection weight of the neuron of hidden layer,It is network input layer to the threshold value of the neuron of the hidden layer,For The hidden layer to the neuron of network output layer connection weight,It is the hidden layer to the nerve of the network output layer The threshold value of unit, wherein, i=1,2 ... S₀；J=1,2 ... S₁；K=1,2 ... S₂；S₀It is the number of the neuron of the network input layer Amount, S₁It is the quantity of the neuron of the network hidden layer, S₂It is the quantity of the neuron of the network output layer；

The original state variable is：

$X={\left[\begin{array}{cccccccc}{w}_{11}^{1}L& w_{s_0{s}_1}^1& b_1^{1}L& b_{s_1}^1& w_{11}^{2}L& w_{s_1{s}_2}^2& b_1^{2}L& b_{s_2}^2\end{array}\right]}^T---\left(2\right)$

Step S5：The optimum state variable of the neural network model is estimated using ST-UKFNN algorithms；

Step S6：Using the optimum state variable as the neural network modelWithFormula (1) is carried out Update, obtain the neural network model after the training sample updates；

Step S7：H is built respectively₂S concentrationPreference function and CO₂ConcentrationPreference function；

Step S8：Using MOGA algorithms respectively to H₂S concentrationPreference function and CO₂ConcentrationPreference function carry out it is many The optimizing of target extreme value optimizes, and acquisition meets the decision variable of produce reality；

Step S9：The neural network model that decision variable after optimization is brought into after the training sample updates, after calculation optimization Decision variable systematic function, the average value with the systematic function of actual sample is compared, if the decision-making after optimization becomes The systematic function of amount is carried out using the decision variable after optimization more than the average value of the systematic function of actual sample to actual production Instruct；Otherwise repeat the above steps S1-S8, until the systematicness of the systematic function more than actual sample of the decision variable after optimization Untill the average value of energy.

2. it is as claimed in claim 1 to purify Technology Modeling and optimization method towards energy-saving high sulfur-containing natural gas, it is described Step S5 includes：

Step S51：Sigma samplings are carried out to the original state variable X, 2n+1 sampled point, initialization control 2n+1 is obtained The distribution parameter alpha of individual sampled point, parameter κ to be selected, and non-negative right factor beta, to the Sigma of the original state variable X Sampling is as follows：

$\left\{\begin{array}{c}\begin{array}{cc}{\hat{X}}_{k-1|k-1}^{\left(i\right)}=X_{k-1}+\sqrt{\left(n+\mathrm{\λ}\right)p_{k-1}}& i=1:n\\ {\hat{X}}_{k-1|k-1}^{\left(i\right)}=X_{k-1}-\sqrt{\left(n+\mathrm{\λ}\right)p_{k-1}}& i=n+1:2n\end{array}\\ \mathrm{\λ}=a^2\left(n+\mathrm{\κ}\right)-n\end{array}\right.---\left(3\right)$

Wherein,For the i-th row that the optimum state variable at (k-1) moment is estimated, n is state matrix dimension, p_k-1It is (k- 1) covariance of the optimum state variable at moment；

Step S52：The weight of each sampled point is calculated, the weight of each sampled point is as follows：

$\left\{\begin{array}{c}{W}_m^{\left(0\right)}=\mathrm{\λ}/(n+\mathrm{\λ})\\ W_c^{\left(0\right)}=\mathrm{\λ}/(n+\mathrm{\λ})+(1-{\mathrm{\α}}^2+\mathrm{\β})\\ \begin{array}{cc}{W}_m^{\left(i\right)}=W_c^{\left(i\right)}=\mathrm{\λ}/(2\×(n+\mathrm{\λ}\left)\right)& i=1:2n\end{array}\end{array}\right.---\left(4\right)$

Wherein, W_cTo calculate the weight of the covariance of state variable, W_mWeight during to calculate state estimation and observation prediction,It isFirst row, W_c ⁽⁰⁾It is W_c ⁽ⁱ⁾First row；

Step S53：By the state equation of Discrete time Nonlinear Systems by the optimum state at (k-1) moment of each sampled point The state estimation of variable is transformed to the state estimation of the state variable at k momentAnd by merging the state estimation at k momentVector, obtain the k moment state variable state prior estimateWith covariance P_k|k-1；Wherein,

The state estimationFor：

$X_{k|k-1}^{\left(i\right)}=F\left(X_{k-1|k-1}^{\left(i\right)}\right)+w_k---\left(5\right)$

Wherein, w_kIt is process noise, its covariance matrix Q_kIt is cov (w_k,w_j)=Q_kδ_kj,

The state prior estimateFor：

${\hat{X}}_{k|k-1}=\underset{i=0}{\overset{2n}{\mathrm{\Σ}}}{W}_m^{\left(i\right)}\·X_{k|k-1}^{\left(i\right)}---\left(6\right)$

The covariance P of the state variable_k|k-1For：

$P_{k|k-1}=\underset{i=0}{\overset{2n}{\mathrm{\Σ}}}{W}_c^{\left(i\right)}\·(X_{k|k-1}^{\left(i\right)}-{\hat{X}}_{k|k-1}){(X_{k|k-1}^{\left(i\right)}-{\hat{X}}_{k|k-1})}^T+Q_{K-1}---\left(7\right)$

Step S54：The state estimation of the state variable at k moment is set up by the observational equation of Discrete time Nonlinear Systems With the observation predicted estimate at k momentBetween contact with complete observation prediction, and estimate the k moment observation prediction association side Difference

The average of the observation prediction at the k momentFor：

${\hat{Y}}_{k|k-1}=\underset{i=0}{\overset{2n}{\mathrm{\Σ}}}{W}_m^{\left(i\right)}\·Y_{k|k-1}^{\left(i\right)}+v_k---\left(8\right)$

Wherein,

Wherein, ν_kIt is observation noise, its covariance matrix R_kIt is cov (v_k,v_j)=R_kδ_kj,

The covariance of the observation prediction at the k momentFor：

$P_{y_k}={\mathrm{\λ}}_{k+1}\underset{i=0}{\overset{2n}{\Σ}}{W}_c^{\left(i\right)}\·(Y_{k|k-1}^{\left(i\right)}-{\hat{Y}}_{k|k-1}){(Y_{k|k-1}^{\left(i\right)}-{\hat{Y}}_{k|k-1})}^T+R_k---\left(10\right)$

Wherein, above-mentioned formula λ_k+1, it is fading factor

${\mathrm{\λ}}_{k+1}=\left\{\begin{array}{cc}{\mathrm{\λ}}_0,& {\mathrm{\λ}}_0>1\\ 1,& {\mathrm{\λ}}_0\≤1\end{array}\right.---\left(11\right)$

${\mathrm{\λ}}_0=\frac{tr{N}_{k+1}}{tr{M}_{k+1}}---\left(12\right)$

$M_{k+1}=\underset{i=0}{\overset{2n}{\mathrm{\Σ}}}{W}_c^{\left(i\right)}\·(Y_{k|k-1}^{\left(i\right)}-{\hat{Y}}_{k|k-1}){(Y_{k|k-1}^{\left(i\right)}-{\hat{Y}}_{k|k-1})}^T+R_k---\left(13\right)$

N_k+1=V_k+1-βR_k+1 (14)

Wherein, β is the reduction factor (β >=1)；

$V_{k+1}=\left\{\begin{array}{cc}{e}_k{e}_k^T& k=0\\ \frac{{\mathrm{\ρV}}_k+e_k{e}_k^T}{1+\mathrm{\ρ}}& k\≥1\end{array}\right.---\left(15\right)$

$e_k=Y_{k|k-1}^{\left(i\right)}-\underset{i=0}{\overset{2n}{\mathrm{\Σ}}}{W}_m^{\left(i\right)}\·g\left(\underset{k=1}{\overset{s_2}{\mathrm{\Σ}}}\left(f\left(\underset{j=1}{\overset{S_1}{\mathrm{\Σ}}}{w}_{ji}^1{X}_{k|k-1}^{\left(i\right)}+b_j^1\right)\right)\·w_{kj}^2+b_k^2\right)---\left(16\right)$

Wherein, ρ ∈ (0,1)；

Step S55：Calculate the covariance P between the state variable at k moment and observation prediction_xy,k：

$P_{xy,k}={\mathrm{\λ}}_{k+1}\underset{i=0}{\overset{2n}{\Σ}}{W}_c^{\left(i\right)}\·(X_{k|k-1}^{\left(i\right)}-{\hat{X}}_{k|k-1}){(Y_{k|k-1}^{\left(i\right)}-{\hat{Y}}_{k|k-1})}^T---\left(17\right)$

Step S56：By setting up covariance P_xy,kAnd covarianceRelation, update the k moment state variable state estimation And covariance, obtain the optimum state variable at k moment；

Step S57：The optimum state variable at the k moment of acquisition is substituted into step S51 and re-starts sigma samplings, circulation step S51-S57, obtains the optimum state variable of the neural network model.

3. Technology Modeling and optimization method are purified towards energy-saving high sulfur-containing natural gas as described in claim 2, its In, the covariance P of the state variable set up in step S56_xy,kWith the covariance of observation predictionRelation be：

$K_k=\frac{P_{xy,k}}{P_{y,k}}---\left(18\right)$

Wherein, K_kIt is gain matrix, the state estimation of the optimum state variable for updating the k moment is realized with this and the shape at k moment is updated The covariance P of state variable_k；And,

The state estimation X of the optimum state variable at the k moment after renewal_k|kFor：

$X_{k|k}=X_k={\hat{X}}_{k|k-1}+K_k(Y_k-{\hat{Y}}_{k|k-1})---\left(19\right)$

The covariance P of the state variable at the k moment after renewal_kFor：

P_k=λ_k+1P_k|k-1-K_kP_ykK_k ^T (20)

By the state estimation X of the state variable at the k moment after renewal_kWith covariance P_kAs the optimum state variable at k moment.

4. it is as claimed in claim 1 to purify Technology Modeling and optimization method towards energy-saving high sulfur-containing natural gas, wherein, Step S8 includes：

Step S81：By the technological parameter x before optimization₁-x₁₀Respectively as decision variable, by decision variable P=[x₁ x₂ L x₁₀] fitness function value comparing find optimized individual；Wherein, returned the performance variable function of part maximizing is counter One changes acquisition fitness function is：

$objFun\left(X\right)=\[\begin{array}{cc}{\mathrm{perf}}_c\left(g^{-1}\right({\hat{y}}_1\left(f\left(X\right)\right)\left)\right)& g^{-1}\left({\hat{y}}_2\right(f\left(X\right)\left)\right)\end{array}\]---\left(21\right)$

Wherein, the performance variable function of part maximizing is：

$\hat{Y}\left(X\right)={\left[\begin{array}{cc}{\hat{y}}_1\left(X\right)& {\hat{y}}_2\left(X\right)\end{array}\right]}^T---\left(22\right)$

Step S82：Using decision variable P=[x₁ x₂ L x₁₀] parent population P is built, wherein,

$P=\left\{(x_{1m}^P,x_{2m}^P,...x_{10m}^P)\right|1\≤m\≤K\}---\left(23\right)$

Wherein, K is the individuality in parent population PQuantity；L is the population sample of initialization This quantity, L=50；GEN is maximum genetic algebra, GEN=100；

Step S83：Bound x according to decision variable_i,min≤x_i≤x_i,max(i=1,2, L, 10) initialization father population P；Its In, the process of initialization father population P is：From the poor amine flow quantity x of entrance tail gas absorber₁Span in random value assign GiveFrom the poor amine flow quantity x for entering two-level absorption tower₂Span in random value assign Until the steam consumption x from vapor preheater₁₀Span in random value assign

Step S84：Father population P to initializing carries out first time genetic iteration (GEN=1) to produce population of future generation；To first The father population P of beginningization carries out the process of first time genetic iteration, including：

Step S841：According to the being dominant property that each is solved in the father population P that fitness function inspection is initialized；Wherein, for one Solution i, its grade r_iEqual to 1 plus the number n better than solution i_i, i.e. r_i=n_i+1；

Step S842：In the population P that will be initialized it is all it is individual be layered according to grade ascending order, then by with one it is linear right Answer function pair each individuality distribution one initial adaptive value；

Step S843：The average value of the initial adaptive value of each individuality each grade Nei is calculated, the average value is in each grade The specified adaptive value of each individuality；

Step S844：Standardization distance in any one grade between any two individuality i and j is calculated by formula (25)：

$d_{ij}=\sqrt{\underset{s=1}{\overset{k}{\Σ}}{\left(\frac{f_s^{\left(i\right)}-f_s^{\left(j\right)}}{f_s^{\max }-f_s^{min}}\right)}^2}---\left(25\right)$

Wherein, f_s ^maxAnd f_s ^minIt is k-th maximum and minimum value of object function；

Step S845：Calculated by formula (26) has same grade r with solution i_iEach solution d_ij；

Wherein, α=1, σ_shareIt is default microhabitat radius；

Each individual microhabitat number is the summation of Sharing Function value in the grade：

${\mathrm{nc}}_i=\underset{j=1}{\overset{\mathrm{\μ}\left(r_i\right)}{\Σ}}sh\left(d_{ij}\right)---\left(27\right)$

Wherein, μ (r_i) it is that all grades are r_iNumber of individuals；

Step S846：The specified adaptive value of each individuality is obtained the shared adaptation of each individuality divided by respective microhabitat number Value；

Step S847：Change of scale is done to all individual shared adaptive values in each grade；

Step S848：Ratio selection, single-point intersection, the variation calculating acquisition next generation are carried out to each grade by change of scale Population；

Step S85：GEN=GEN+1, circulates 100 step S83~step S84, obtains GEN for population as optimum results Output；Wherein, optimum results It is the technological parameter x after optimization₁-x₁₀。