CN109058151A - A kind of Control Methods for Surge Prevention in Centrifugal Compressors based on prediction model - Google Patents

Abstract

Translated fromChinese

本发明公开的基于预测模型的离心压缩机防喘振控制方法，通过将离心压缩机分解为两个二入单出模型，并通过最小二乘双支持向量回归机方法优化两个模型，然后构建参考轨迹，并在参考轨迹中加入限制条件，控制器可以提前判断工作点是否越过控制线，使得由参考轨迹所确定的工作点始终位于控制线的右侧，因此可以提前发出控制动作打开回流阀，避免了喘振发生，有效提高了离心压缩机的运行效率。

The anti-surge control method of the centrifugal compressor based on the predictive model disclosed in the present invention decomposes the centrifugal compressor into two two-input single-out models, and optimizes the two models through the least squares double support vector regression machine method, and then constructs The reference trajectory, and adding restrictive conditions in the reference trajectory, the controller can judge in advance whether the operating point crosses the control line, so that the operating point determined by the reference trajectory is always on the right side of the control line, so the control action can be issued in advance to open the return valve , to avoid the occurrence of surge, and effectively improve the operating efficiency of the centrifugal compressor.

Description

A kind of Control Methods for Surge Prevention in Centrifugal Compressors based on prediction model

Technical field

The invention belongs to centrifugal compressor Anti-surge Control technical fields, and in particular to it is a kind of based on prediction model fromHeart compressor anti-surge control method.

Background technique

Petrochemical industry is the important backbone industry of modern age developed country.In recent years, growth of the national economic situation occurredGreat changes, influence of the petrochemical industry to national economy is also increasing, play in the development of Chinese national economyIt is extremely important effect oneself become Chinese national economy basis and development China national industry, keep national economy continue surelySurely the mainstay industry developed.Compressor is the key equipment in petrochemical production equipment, is mainly used for by gasIt is compressed and is conveyed, create necessary condition for chemical reaction；With the progress of world technology, the fast development of petrochemical fieldDemand to centrifugal compressor is increasing, while also proposing higher want to centrifugal compressor manufacturing technology and control systemIt asks；Anti-surge Control is emphasis in centrifugal compressed machine control system, difficult point, and Anti-surge Control is improper seriously to limit centrifugationThe stable operation range and operational efficiency of compressor even can damage centrifugal compressor when serious, make to production and personal safetyAt imponderable loss.How under the premise of meeting technique requirement, it is effectively prevent the generation of surge, while improving as far as possibleThe operational efficiency of centrifugal compressor is always critical issue therein.

Current most popular anti-surge control method is still passive control methods, therefore develops technological innovation, Control Methods for Surge Prevention in Centrifugal Compressors with independent intellectual property rights is for improving centrifugal compressor operational efficiency and guarantorCard centrifugal compressor is stable, continue, health operation has a very important significance.

Summary of the invention

The purpose of the present invention is a kind of Control Methods for Surge Prevention in Centrifugal Compressors based on prediction model, both effectively preventThe generation of centrifugal compressor surge, while also improving the operational efficiency of centrifugal compressor.

The technical scheme adopted by the invention is that a kind of Control Methods for Surge Prevention in Centrifugal Compressors based on prediction model,Specifically implement in accordance with the following steps:

Step 1, the operation data for acquiring centrifugal compressor system, and using the operation data as sample set, and to sampleThis collection is pre-processed；

Step 2, the prediction model for tentatively establishing centrifugal compressor；

Step 3, the prediction model by determining Model Parameter Optimization centrifugal compressor；

Step 4, by optimization after centrifugal compressor prediction model construct reference locus, the reference locus be from fromHeart compressor control system is currently practical to set out, to the curve of setting value transition；

The reference locus of control line restrictive condition is added in step 5, building；

Step 6 carries out feedback compensation to the reference locus in step 5, obtains prediction correcting model；

Step 7 carries out rolling optimization by control system output valve of the prediction correcting model to centrifugal compressor, obtains KThe outlet pressure of moment centrifugal compressor and the performance indicator of mass flow and control constraints condition.

The features of the present invention also characterized in that

In step 1, the collection process of the operation data of centrifugal compressor system is as follows:

Outlet pressure, mass flow, revolving speed and the outlet throttling valve opening of centrifugal compressor are recorded, and as sample set；Then the dissimilarity concentrated according to criterion Rejection of samples finally obtains 300 sample points as training set；In the control of centrifugal compressorInterference protection measure is added in processing procedure sequence, the dissimilarity then concentrated according to criterion Rejection of samples finally obtains 200 sample point conductsTest set.

In step 1, pretreatment is normalized to the sample set, processing mode is as follows:

In formula (1), x_iValue is actually entered for control system.

Detailed process is as follows for step 2:

Step 2.1, establish the two of centrifugal compressor enter it is single go out model, including outlet pressure model and mass flow mouldType is as follows respectively:

M (t+1)=f_m(m(t),m(t-1),P_d(t-2),u₁(t),...,u₁(t-m_1,m),u₂(t),...,u₂(t-m_2,m)) (3)；

In formula (2) and formula (3), P_dIt is the outlet pressure of centrifugal compressor, m is the mass flow of centrifugal compressor, controlInput u₁And u₂It is revolving speed and outlet throttling valve opening, m respectively₁And m₂It is the order of control input；

Step 2.2, to the m in each model₁And m₂Value be combined, every kind of combination is established respectively with training setThen model is verified with test set, find out the root-mean-square error RMSE of every kind of established model of combination:

In formula (4), y* is model output value, y_iIt is system real output value, n is test sample number, and taking n is 200；

The prediction model of step 2.3, centrifugal compressor when taking root-mean-square error RMSE minimum:

M (t+1)=f_m(m(t),m(t-1),P_d(t-2),u₁(t),u₁(t-1),u₂(t)) (6)；

In step 3, the optimization process of the prediction model of the centrifugal compressor is as follows:

Step 3.1, determine centrifugal compressor prediction model model parameter；

Step 3.2 after determining model parameter, introduces regression vector P in outlet pressure and the model of mass flow respectively_dAnd m, it obtains such as minor function:

In formula (7) and formula (8),w_mFor with reference to coefficient, b_m、For amount of bias；

Step 3.2 solves above-mentioned function, obtains the model of outlet pressure and mass flow, form are as follows:

In formula (9) and formula (10),For Lagrange duality variable,And k_m(m,m_i) it is core letterNumber.

In step 3.1, the determination process of the model parameter is as follows:

(1) setting model parameter C and σ²Candidate Set be more open grid: { 2^-5,2^-3,...,2¹⁵And { 2^-9,2^-7,...,2¹¹}；

(2) times sample cross is carried out as parameter sample using the node in grid to examine, obtain maximum crosscheck precision,Corresponding grid node outlet pressure model is { 2^-3,2^-5, mass flow model is { 2^-3,2^-3}；

(3) centered on the mesh point for obtaining maximum crosscheck precision, thinner grid is constructed in a certain range:Outlet pressure model meshes are { 2^-5,2^-4.5,...,2^-1And { 2^-7,2^-6.5,...,2^-3, mass flow model meshes are as follows: { 2^-5,2^-4.5,...,2^-1And { 2^-7,2^-4.5,...,2^-1}；

(4) again using the node in grid as parameter sample, and 10 times of sample cross are carried out to it and are examined, finally obtain inspectionTest the C and σ of the highest outlet pressure model of precision²Value is { 2^-3,2^-6, the C and σ of mass flow model²Value is { 2^-2.5,2^-4.5}。

In step 4, the reference locus of the outlet pressure and mass flow is respectively as follows:

In formula (13) and formula (14), S_PdAnd S_mIt is the setting value of outlet pressure and mass flow, P respectively_d(k) divide with m (k)It is not the reality output of k moment outlet pressure and mass flow, P_{D, r}(k+i) it is exported for the reference of outlet pressure, m_r(k+i) it isThe reference of mass flow exports, T_ref,PdAnd T_ref,mIt is the time constant of outlet pressure and mass flow reference locus respectively.

In step 5, the reference locus of control line restrictive condition is added are as follows:

In formula (16), m_CL(P_d,r(k+i)) it indicates on the control line, when outlet pressure is P_d,r(k+i) mass flow when,P_d,CL(m_r(k+i)) it indicates on the control line, when mass flow is m_r(k+i) outlet pressure when, restrictive condition P_d,r(k+i)≤P_d,CL(m_r(k+i)) indicate reference locus determined by operating point be located on control line or control line right side, restrictive condition P_d,r(k+I) > P_d,CL(m_r(k+i)) indicate that operating point determined by reference locus is located on the left of control line.

In step 6, the prediction correcting model is as follows:

m_p(k+i)=m_mp(k+i)+k_m(m(k)-m_mp(k)) (18)；

In formula (17) and formula (18), k is Ratio for error modification, and d is prediction time domain；P_d(k) and m (k) is currently to be respectivelyThe practical outlet pressure of system and mass flow.

In step 7, the outlet pressure of the centrifugal compressor and the performance indicator difference of mass flow are as follows:

In formula (19) and formula (20),It is the weighting coefficient for predicting output error, j >=0 λ is the weighting coefficient of control amount, dIt is prediction time domain, Nu controls time domain, and d >=Nu >=1, v are the control output of centrifugal compressor revolving speed, k_tIt is that speed control muffler control is defeatedOut；

The control constraints condition is as follows:

s.t. v_min≤x≤v_max(22),

0≤k_t≤ 100 (23),

m_min≤m_p(k+i)≤m_max(24),

P_{D, p}(k+i)≤P_{D, CL}(m_p(k+i)) (26)；

In formula (21)~(26), V_minAnd V_maxThe respectively minimum and maximum revolving speed of centrifugal compressor；m_minAnd m_maxRespectivelyIt is centrifugal compressor minimum and maximum mass flow, P_d,minAnd P_d,maxIt is centrifugal compressor minimum and maximum outlet pressure respectively,P_d,CL(m_p(k+i)) it indicates on the control line when mass flow is m_p(k+i) corresponding outlet pressure when；Formula (22) and formula(23) be actuator itself constraint condition, formula (24) and formula (25) are the constraint condition of centrifugal compressor working range, formulaIt (26) is anti-surge constraint condition.

The beneficial effects of the present invention are:

Control Methods for Surge Prevention in Centrifugal Compressors based on prediction model of the invention, by the way that centrifugal compressor to be decomposed intoTwo two enter single model out, and optimize two models by the double support vector regression methods of least square, and then building refers toTrack, and restrictive condition is added in reference locus, controller can judge in advance whether operating point crosses control line, so that byOperating point determined by reference locus is always positioned at the right side of control line, therefore can issue control action in advance and open refluxValve avoids surge, effectively increases the operational efficiency of centrifugal compressor.

Detailed description of the invention

Fig. 1 is a kind of control flow chart of the Control Methods for Surge Prevention in Centrifugal Compressors based on prediction model of the present invention；

Fig. 2 is a kind of Anti-surge Control curve of the Control Methods for Surge Prevention in Centrifugal Compressors based on prediction model of the present inventionFigure；

Fig. 3 is the centrifugal compressor in a kind of Control Methods for Surge Prevention in Centrifugal Compressors based on prediction model of the present inventionSurge line；

Fig. 4 is the centrifugal compressor in a kind of Control Methods for Surge Prevention in Centrifugal Compressors based on prediction model of the present inventionSurge curve and control line；

Fig. 5 is to carry out centrifugal compressed using a kind of Control Methods for Surge Prevention in Centrifugal Compressors based on prediction model of the present inventionThe control analogous diagram of the operating point of machine.

Specific embodiment

The following describes the present invention in detail with reference to the accompanying drawings and specific embodiments.

A kind of Control Methods for Surge Prevention in Centrifugal Compressors based on prediction model of the present invention, as shown in Figure 1, specifically according to such asLower step is implemented:

Step 1, the operation data for acquiring centrifugal compressor system, and using the operation data as sample set, and to sampleThis collection is pre-processed；

In order to obtain the sample set for training and testing prediction model, it is necessary to acquire data and be located in advance to dataReason: during system trial run, by the operation data of configuration software recording compressed machine system, the sampling time is the second.Stop controlOutlet pressure controllers and mass flow controller in system processed, by changing the revolving speed of driver and opening for speed control mufflerDegree allows the operating point of centrifugal compressor to change on the right side of its surge line, while recording the outlet pressure of centrifugal compressor, quality streamAmount, revolving speed and outlet throttling valve opening, and as sample set；For training sample, any measure is not taken directly to record itData finally obtain 300 sample points as training set according to the dissimilarity that criterion Rejection of samples is concentrated；For test sample, it isIncrease reliability, in the control program of centrifugal compressor is added interference protection measure, then concentrates according to criterion Rejection of samplesDissimilarity, finally obtain 200 sample points as test set.

Since each input variable has different physical significances, variation range is different, it is therefore desirable to sample setPretreatment is normalized, processing mode is as follows:

In formula (1), x_iValue is actually entered for control system.

Step 2, the prediction model for tentatively establishing centrifugal compressor；

Detailed process is as follows:

Step 2.1, the present invention are using revolving speed control outlet pressure, the control strategy of outlet throttling control mass flow, thenCentrifugal compressor is decomposed into two two and enters single model, respectively outlet pressure model and mass flow model out, the two modelsForm are as follows:

M (t+1)=f_m(m(t),m(t-1),P_d(t-2),u₁(t),...,u₁(t-m_1,m),u₂(t),...,u₂(t-m_2,m)) (3)；

In formula (2) and formula (3), P_dIt is the outlet pressure of centrifugal compressor, m is the mass flow of centrifugal compressor, controlInput u₁And u₂It is revolving speed and outlet throttling valve opening, m respectively₁And m₂It is the order of control input.

Step 2.2, the order in order to determine control input, by the m of each model₁And m₂All possible value carries out groupIt closes, to the m in each model₁And m₂Value be combined, every kind of combination is respectively established with training set, then with surveyExamination collection is verified, and the root-mean-square error RMSE of every kind of established model of combination is found out:

In formula (4), y* is model output value, y_iIt is system real output value, n is test sample number, and taking n is 200；

The prediction model of step 2.3, centrifugal compressor when taking root-mean-square error RMSE minimum:

M (t+1)=f_m(m(t),m(t-1),P_d(t-2),u₁(t),u₁(t-1),u₂(t)) (6)。

Step 3, the prediction model by determining Model Parameter Optimization centrifugal compressor；

Detailed process is as follows:

Step 3.1, determine centrifugal compressor prediction model model parameter；

(1) setting model parameter C and σ²Candidate Set be more open grid: { 2^-5,2^-3,...,2¹⁵And { 2^-9,2^-7,...,2¹¹}；

(2) times sample cross is carried out as parameter sample using the node in grid to examine, obtain maximum crosscheck precision,Corresponding grid node outlet pressure model is { 2^-3,2^-5, mass flow model is { 2^-3,2^-3}；

(3) centered on the mesh point for obtaining maximum crosscheck precision, thinner grid is constructed in a certain range:Outlet pressure model meshes are { 2^-5,2^-4.5,...,2^-1And { 2^-7,2^-6.5,...,2^-3, mass flow model meshes are as follows: { 2^-5,2^-4.5,...,2^-1And { 2^-7,2^-4.5,...,2^-1}；

(4) again using the node in grid as parameter sample, and 10 times of sample cross are carried out to it and are examined, finally obtain inspectionTest the C and σ of the highest outlet pressure model of precision²Value is { 2^-3,2^-6, the C and σ of mass flow model²Value is { 2^-2.5,2^-4.5}。

Step 3.2 after determining model parameter, introduces regression vector P in outlet pressure and the model of mass flow respectively_dAnd m,

Wherein, for outlet pressure model, regression vector are as follows:

P_d,i=(P_d(i),P_d(i-1)P_d(i-2),u₁(i),u₁(i-1),u₂(i)) (27),

Its sample set are as follows: { P_d(i+1),P_d,i, i=1,2 ..., l；

For mass flow model, regression vector is

m_i=(m (i), m (i-1), m (i-2), u₁(i),u₁(i-1),u₂(i)) (28),

Its sample set are as follows: { m (i+1), m_i, i=1,2 ..., l；

It is to obtain wait find a function such as minor function by model conversation:

In formula (7) and formula (8),w_mFor with reference to coefficient, b_m、For amount of bias；

Step 3.2 solves above-mentioned function, obtains the model of outlet pressure and mass flow, form are as follows:

In formula (9) and formula (10), α_m,i、For Lagrange duality variable,And k_m(m,m_i) it is core letterNumber.

Step 4 constructs reference locus by the prediction model of the centrifugal compressor after optimization, and reference locus is to press from centrifugationContracting machine control system is currently practical to set out, to the curve of setting value transition；

In Model Algorithmic contral, the desired output of control system is by from the currently practical output of system, to setting valueAs defined in the reference locus to smoothly transit；The k moment reference locus can by its following sampling instant value y_rTo retouchIt states, it usually may be taken as the form of first order exponential variation.

Outlet pressure and the reference locus of mass flow are respectively as follows:

In formula (13) and formula (14), S_PdAnd S_mIt is the setting value of outlet pressure and mass flow, P respectively_d(k) divide with m (k)It is not the reality output of k moment outlet pressure and mass flow, P_{D, r}(k+i) it is exported for the reference of outlet pressure, m_r(k+i) it isThe reference of mass flow exports, T_ref,PdAnd T_ref,mIt is the time constant of outlet pressure and mass flow reference locus respectively.

The reference locus of control line restrictive condition is added in step 5, building, so that the operating point determined by reference locus is begunFinal position is in the right side of control line, as shown in Figure 2.

Reference locus is the guidance to controller, and controller issues control action under the guidance of reference locus, to makeSystem output changes along reference locus.Therefore in the ginseng of centrifugal compressor anti-surge PREDICTIVE CONTROL middle outlet pressure and mass flowThe restrictive condition that control line is added in track is examined, the reference locus of control line restrictive condition is added are as follows:

In formula (16), m_CL(P_d,r(k+i)) it indicates on the control line, when outlet pressure is P_d,r(k+i) mass flow when,P_d,CL(m_r(k+i)) it indicates on the control line, when mass flow is m_r(k+i) outlet pressure when, restrictive condition P_d,r(k+i)≤P_d,CL(m_r(k+i)) indicate reference locus determined by operating point be located on control line or control line right side, restrictive condition P_d,r(k+I) > P_d,CL(m_r(k+i)) indicate that operating point determined by reference locus is located on the left of control line.By being added in reference locusRestrictive condition, it can be ensured that the operating point determined by reference locus is always positioned at the right side of control line.

Step 6 carries out feedback compensation to the reference locus in step 5, obtains prediction correcting model；

Model prediction output and system certainly exist error between really exporting, and to guarantee to control precision, need to errorFeedback compensation is carried out, the prediction correcting model after feedback is added is as follows:

m_p(k+i)=m_mp(k+i)+k_m(m(k)-m_mp(k)) (18)；

In formula (17) and formula (18), k is Ratio for error modification, and d is prediction time domain；P_d(k) and m (k) is currently to be respectivelyThe practical outlet pressure of system and mass flow.

Step 7 carries out rolling optimization by control system output valve of the prediction correcting model to centrifugal compressor, obtains KThe outlet pressure of moment centrifugal compressor and the performance indicator of mass flow and control constraints condition.

Inside model algorithm, the Optimality Criteria at k moment is will be according to the following Nu control amount, so that future d momentPrediction output is as close as desired output determined by reference locus.

Rolling optimization exports energy with the smallest controller and obtains optimal control effect.

K moment, the outlet pressure of the centrifugal compressor and the performance indicator difference of mass flow are as follows:

In formula (19) and formula (20),It is the weighting coefficient for predicting output error, j >=0 λ is the weighting coefficient of control amount, dIt is prediction time domain, Nu controls time domain, and d >=Nu >=1, v are the control output of centrifugal compressor revolving speed, k_tIt is that speed control muffler control is defeatedOut；

In view of the limitation of centrifugal compressor revolving speed, throttle valve opening, working range and control line, Anti-surge Control is askedTopic can be converted into the optimization problem of following control constraints condition:

s.t. v_min≤v≤v_max(22),

0≤k_t≤ 100 (23),

m_min≤m_p(k+i)≤m_max(24),

P_{D, p}(k+i)≤P_{D, CL}(m_p(k+i)) (26)；

In formula (21)~(26), V_minAnd V_maxThe respectively minimum and maximum revolving speed of centrifugal compressor；m_minAnd m_maxRespectivelyIt is centrifugal compressor minimum and maximum mass flow, P_d,minAnd P_d,maxIt is centrifugal compressor minimum and maximum outlet pressure respectively,P_d,CL(m_p(k+i)) it indicates on the control line when mass flow is m_p(k+i) corresponding outlet pressure when；Formula (22) and formula(23) be actuator itself constraint condition, formula (24) and formula (25) are the constraint condition of centrifugal compressor working range, formulaIt (26) is anti-surge constraint condition.

The present invention solves above-mentioned control constraints condition using MOUSE algorithm.This algorithm is similar to genetic algorithm,It is a kind of global random searching optimizing algorithm, but it does not generate filial generation by intersecting and making a variation, each filial generation is by only changingThe conventional search optimizing algorithm of one variable generates.The complexity of algorithm greatly reduces, and convergence rate greatly improves, energyEnough meet the requirement of control system real-time.

The present invention, using centrifugal compressor characteristic as the function of revolving speed, is controlled to model conveniently by adjusting revolving speedPressure lifting, mass flow are controlled by adjusting throttle valve opening change throttle valve resistance coefficient gas；Centrifugal compressor characteristicIt is obtained by experimental method.In order to protect centrifugal compressor, the present invention carries out anti-surge control using improved Greitzer modelSystem emulation, model form are as follows:

In formula (29) and formula (30), m is centrifugal compressor mass flow, and p is container pressure, p₀₁It is inlet pressure, a₀₁It isEntrance stagnation velocity of sound, Vp are container volume, L_cIt is centrifugal compressor and duct length, A₁It is area of reference, k_tIt is throttle valve resistanceCoefficient, ψ_cIt is centrifugal compressor characteristic, it is the nonlinear function of centrifugal compressor mass flow and revolving speed, can be obtained by experiment?.

According to the point tested on resulting centrifugal compressor characteristic curve, while considering flow when centrifugal compressor surgeIt may be negative, centrifugal compressor performance curve is obtained using curve-fitting method:

(1) the crucial revolving speed for choosing several centrifugal compressors, takes rated speed, maximum (top) speed, minimum speed, volume hereinDetermine the crucial revolving speed of revolving speed, rated speed this 5；

(2) by changing outlet throttling valve opening under each crucial revolving speed, while paying attention to not allowing operating point to cross surgeLine obtains the relation curve of mass flow and outlet pressure, and here it is the performance curves of centrifugal compressor, and on every curveUniformly take 5 points；

(3) method for using curve matching, while considering that flow may be negative when surge occurs for centrifugal compressor, utilizeCubic curve simulates the performance curve of centrifugal compressor；

(4) point that performance curve upper outlet pressure obtains maximum value is connected, has just obtained the asthma of centrifugal compressorShake line, as shown in Figure 3.

The prediction time domain and control time domain of PREDICTIVE CONTROL are bigger, and the information about system future input and output is abundanter,Control effect is better, it is contemplated that calculation amount, prediction time domain and control time domain can not be too big, and the present invention takes control time domain N_u=3, predict that time domain d=4, each weight in performance indicator are taken as 0.5.

Since PREDICTIVE CONTROL can issue control action according to the output of prediction model in advance, actuator can reduceThe influence of regulating time, the smallest Con trolling index of two secondary controls input difference can reduce operating point fluctuation in addition, therefore can incite somebody to actionThe distance between control line and surge line of centrifugal compressor greatly shorten, and control line are set in away from the right side of surge line when emulationAt 5%, as shown in Figure 4.

Anti-surge actuator uses return valve, and valve opening signal joined 2 in order to simulate the characteristic of valve in practiceThe delay of second, shutdown signal joined delay in 8 seconds.

It is emulated first using variable limit flow method, changes the operating point of centrifugal compressor by C point in Fig. 5 to D point,As can be seen from Figure 5, new operating point is closer away from surge line, when operating point is changed near D point, due to controller overshoot etc. becauseThe influence of element, operating point are continued mobile to cross control line to the left.Since variable limit flow method can not judge in advanceThe running track of operating point, therefore only when control line is crossed in operating point, controller can just issue control action and open refluxValve, but due to the lag characteristic that return valve is opened, cause operating point not return in time on the right side of control line, and be to continue with to the leftIt has been moved across surge line, therefore has resulted in centrifugal compressor and surge has occurred.

It is emulated using the Control Methods for Surge Prevention in Centrifugal Compressors proposed by the present invention based on prediction model, by Fig. 5In as it can be seen that change centrifugal compressor operating point by C point in figure to D point, the centrifugal compressor anti-surge control based on prediction modelMethod processed, controller can judge whether operating point crosses control line in advance, therefore can issue in advance control action and open backValve is flowed, avoids surge, and overshoot is small, operating point variation is steady, makes centrifugal compressor steady operation in D point.

Claims (10)

Translated fromChinese

1.一种基于预测模型的离心压缩机防喘振控制方法，其特征在于，具体按照如下步骤实施：1. A centrifugal compressor anti-surge control method based on predictive model, is characterized in that, specifically implements according to the following steps:步骤1、采集离心压缩机系统的运行数据，并将所述运行数据作为样本集，并对样本集进行预处理；Step 1, collecting the operation data of the centrifugal compressor system, using the operation data as a sample set, and preprocessing the sample set;步骤2、初步建立离心压缩机的预测模型；Step 2. Preliminary establishment of a prediction model for the centrifugal compressor;步骤3、通过确定模型参数优化离心压缩机的预测模型；Step 3, optimizing the predictive model of the centrifugal compressor by determining the model parameters;步骤4、通过优化后的离心压缩机的预测模型构建参考轨迹，所述参考轨迹为从离心压缩机控制系统当前实际输出出发,向设定值过渡的曲线；Step 4, constructing a reference trajectory through the prediction model of the optimized centrifugal compressor, the reference trajectory is a curve transitioning from the current actual output of the centrifugal compressor control system to a set value;步骤5、构建加入控制线限制条件的参考轨迹；Step 5, constructing the reference trajectory that joins the control line restriction;步骤6、对步骤5中的参考轨迹进行反馈校正，得到校正预测模型；Step 6, performing feedback correction on the reference trajectory in step 5 to obtain a correction prediction model;步骤7、通过校正预测模型对离心压缩机的控制系统输出值进行滚动优化，得到K时刻离心压缩机的出口压力和质量流量的性能指标及控制约束条件。Step 7. Carry out rolling optimization on the output value of the control system of the centrifugal compressor by correcting the prediction model, and obtain the outlet pressure and mass flow performance indicators and control constraints of the centrifugal compressor at time K.2.如权利要求1所述的一种基于预测模型的离心压缩机防喘振控制方法，其特征在于，步骤1中，离心压缩机系统的运行数据的采集过程如下：2. a kind of anti-surge control method of centrifugal compressor based on predictive model as claimed in claim 1, is characterized in that, in step 1, the collection process of the operation data of centrifugal compressor system is as follows:记录离心压缩机的出口压力、质量流量、转速及出口节流阀开度，并作为样本集；然后依据准则剔除样本集中的异点,最后获得300个样本点作为训练集；在离心压缩机的控制程序中加入抗干扰措施，然后依据准则剔除样本集中的异点，最后取得200个样本点作为测试集。Record the outlet pressure, mass flow, speed and outlet throttle valve opening of the centrifugal compressor, and use it as a sample set; then eliminate the outliers in the sample set according to the criteria, and finally obtain 300 sample points as a training set; Anti-interference measures are added to the control program, and then the outliers in the sample set are eliminated according to the criteria, and finally 200 sample points are obtained as the test set.3.如权利要求2所述的一种基于预测模型的离心压缩机防喘振控制方法，其特征在于，步骤1中，对所述样本集进行归一化预处理，处理方式如下：3. A kind of anti-surge control method of centrifugal compressor based on predictive model as claimed in claim 2, it is characterized in that, in step 1, described sample set is carried out normalized preprocessing, and processing mode is as follows:式(1)中，x_i为控制系统实际输入值。In formula (1), x_i is the actual input value of the control system.4.如权利要求3所述的一种基于预测模型的离心压缩机防喘振控制方法，其特征在于，步骤2的具体过程如下：4. A kind of anti-surge control method of centrifugal compressor based on predictive model as claimed in claim 3, is characterized in that, the specific process of step 2 is as follows:步骤2.1、建立离心压缩机的二入单出模型，其中包括出口压力模型和质量流量模型，分别如下：Step 2.1, establish the two-input single-outlet model of the centrifugal compressor, which includes the outlet pressure model and the mass flow model, respectively as follows:m(t+1)＝f_m(m(t),m(t-1),P_d(t-2),u₁(t),...,u₁(t-m_1,m),u₂(t),...,u₂(t-m_2,m)) (3)；m(t+1)=f_m (m(t),m(t-1),P_d (t-2),u₁ (t),...,u₁ (tm_1,m ),u₂ (t),...,u₂ (tm_2,m )) (3);式(2)和式(3)中，P_d是离心压缩机的出口压力，m是离心压缩机的质量流量，控制输入u₁和u₂分别是转速和出口节流阀开度，m₁和m₂是控制输入的阶次；In formulas (2) and (3), P_d is the outlet pressure of the centrifugal compressor, m is the mass flow rate of the centrifugal compressor, and the control inputs u₁ and u₂ are the speed and outlet throttle valve opening respectively, m₁ and_m2 is the order of the control input;步骤2.2、对每个模型中的m₁和m₂的取值进行组合，用训练集对每种组合分别建立模型，然后用测试集进行验证，求出每种组合所建立的模型的均方根误差RMSE：Step 2.2, combine the values of_m1 and_m2 in each model, use the training set to build a model for each combination, and then use the test set to verify, and find the mean square of the model established by each combination Root error RMSE:式(4)中，y*为模型输出值，y_i是系统实际输出值，n为测试样本个数，取n为200；In formula (4), y* is the output value of the model, y_i is the actual output value of the system, n is the number of test samples, take n as 200;步骤2.3、取均方根误差RMSE最小时的离心压缩机的预测模型：Step 2.3, take the prediction model of the centrifugal compressor when the root mean square error RMSE is the smallest:m(t+1)＝f_m(m(t),m(t-1),P_d(t-2),u₁(t),u₁(t-1),u₂(t)) (6)。m(t+1)＝f_m (m(t),m(t-1),P_d (t-2),u₁ (t),u₁ (t-1),u₂ (t)) (6).5.如权利要求4所述的一种基于预测模型的离心压缩机防喘振控制方法，其特征在于，步骤3中，所述离心压缩机的预测模型的优化过程如下：5. a kind of anti-surge control method of centrifugal compressor based on predictive model as claimed in claim 4, is characterized in that, in step 3, the optimization process of the predictive model of described centrifugal compressor is as follows:步骤3.1、确定离心压缩机的预测模型的模型参数；Step 3.1, determining the model parameters of the prediction model of the centrifugal compressor;步骤3.2、确定模型参数后，在出口压力和质量流量的模型中分别引入回归向量P_d和m，Step 3.2, after determining the model parameters, the regression vectors P_d and m are respectively introduced into the outlet pressure and mass flow models,得到如下函数：Get the following function:式(7)和式(8)中，w_m为参考系数，b_m、为偏置量；In formula (7) and formula (8), w_m is the reference coefficient, b_m , is the offset;步骤3.2、求解上述函数，得到出口压力和质量流量的模型，其形式为：Step 3.2, solve the above-mentioned function, obtain the model of outlet pressure and mass flow rate, its form is:式(9)和式(10)中，α_m,i、为拉格朗日对偶变量，和k_m(m,m_i)为核函数。In formula (9) and formula (10), α_m,i , is the Lagrangian dual variable, and k_m (m,m_i ) are kernel functions.6.如权利要求5所述的一种基于预测模型的离心压缩机防喘振控制方法，其特征在于，步骤3.1中，所述模型参数的确定过程如下：6. A kind of predictive model-based centrifugal compressor anti-surge control method as claimed in claim 5, is characterized in that, in step 3.1, the determination process of described model parameter is as follows:(1)设定模型参数C和σ²的候选集为比较松散的网格：{2^-5,2^-3,...,2¹⁵}和{2^-9,2^-7,...,2¹¹}；(1) Set the candidate sets of model parameters C and σ² as relatively loose grids: {2^-5 ,2^-3 ,...,2¹⁵ } and {2^-9 ,2^-7 ,... ,2¹¹ };(2)以网格中的节点为参数样本进行倍样本交叉检验,得到最大的交叉检验精度，所对应的网格节点出口压力模型为{2^-3,2^-5}，质量流量模型为{2^-3,2^-3}；(2) Take the nodes in the grid as parameter samples to carry out double-sample cross-check to obtain the maximum cross-check accuracy. The corresponding grid node outlet pressure model is {2^-3 ,2^-5 }, and the mass flow model is {^2-3 ,^2-3 };(3)以获得最大交叉检验精度的网格点为中心，在一定范围内构造比较细的网格：出口压力模型网格为{2^-5,2^-4.5,...,2^-1}和{2^-7,2^-6.5,...,2^-3}，质量流量模型网格为：{2^-5,2^-4.5,...,2^-1}和{2^-7,2^-4.5,...,2^-1}；(3) To obtain the grid point with the maximum cross-check accuracy as the center, construct a finer grid within a certain range: the outlet pressure model grid is {2^-5 ,2^-4.5 ,...,2^-1 } and {2^-7 ,2^-6.5 ,...,2^-3 }, the mass flow model meshes are: {2^-5 ,2^-4.5 ,...,2^-1 } and {2^-7 ,2^-4.5 ,...,2^-1 };(4)再次以网格中的节点为参数样本，并对其进行10倍样本交叉检验,最后得到检验精度最高的出口压力模型的C和σ²值为{2^-3,2^-6}，质量流量模型的C和σ²值为{2^-2.5,2^-4.5}。(4) Take the nodes in the grid as parameter samples again, and perform 10-fold sample cross-check on them, and finally obtain the C and σ² values of the outlet pressure model with the highest test accuracy {2^-3 ,2^-6 }, The C and σ² values of the mass flow model are {2^-2.5 ,2^-4.5 }.7.如权利要求5所述的一种基于预测模型的离心压缩机防喘振控制方法，其特征在于，步骤4中，所述出口压力和质量流量的参考轨迹分别为：7. A kind of predictive model-based centrifugal compressor anti-surge control method as claimed in claim 5, is characterized in that, in step 4, the reference track of described outlet pressure and mass flow is respectively:式(13)和式(14)中，S_Pd和S_m分别是出口压力和质量流量的设定值，P_d(k)和m(k)分别是k时刻出口压力和质量流量的实际输出，P_d，r(k+i)为出口压力的参考输出，m_r(k+i)为质量流量的参考输出，T_ref,Pd和T_ref,m分别是出口压力和质量流量参考轨迹的时间常数。In formula (13) and formula (14), S_Pd and S_m are the set values of outlet pressure and mass flow rate respectively, and P_d (k) and m(k) are the actual output of outlet pressure and mass flow rate at time k, respectively , P_{d, r} (k+i) is the reference output of the outlet pressure, m_r (k+i) is the reference output of the mass flow rate, T_ref,Pd and T_ref,m are the reference tracks of the outlet pressure and mass flow rate respectively time constant.8.如权利要求7所述的一种基于预测模型的离心压缩机防喘振控制方法，其特征在于，步骤5中，加入控制线限制条件的参考轨迹为：8. A kind of anti-surge control method of centrifugal compressor based on predictive model as claimed in claim 7, it is characterized in that, in step 5, the reference track of adding control line restriction condition is:式(16)中，m_CL(P_d,r(k+i))表示在控制线上，当出口压力为P_d,r(k+i)时的质量流量，P_d,CL(m_r(k+i))表示在控制线上，当质量流量为m_r(k+i)时的出口压力，限制条件P_d,r(k+i)≤P_d,CL(m_r(k+i))表示参考轨迹所确定的工作点位于控制线上或控制线右侧，限制条件P_d,r(k+i)＞P_d,CL(m_r(k+i))表示参考轨迹所确定的工作点位于控制线左侧。In formula (16), m_CL (P_d,r (k+i)) represents the mass flow rate on the control line when the outlet pressure is P_d,r (k+i), and P_d,CL (m_r (k+i)) represents the outlet pressure on the control line when the mass flow rate is m_r (k+i), and the restriction condition P_d,r (k+i)≤P_d,CL (m_r (k+ i)) indicates that the working point determined by the reference trajectory is located on the control line or on the right side of the control line, and the constraint condition P_d,r (k+i)＞P_d,CL (m_r (k+i)) indicates that the reference trajectory The determined operating point is located to the left of the control line.9.如权利要求8所述的一种基于预测模型的离心压缩机防喘振控制方法，其特征在于，步骤6中，所述校正预测模型如下：9. A kind of anti-surge control method of centrifugal compressor based on predictive model as claimed in claim 8, is characterized in that, in step 6, described correction predictive model is as follows:m_p(k+i)＝m_mp(k+i)+k_m(m(k)-m_mp(k)) (18)；m_p (k+i)=m_mp (k+i)+k_m (m(k)-m_mp (k)) (18);式(17)和式(18)中，k是误差修正系数，d是预测时域；P_d(k)和m(k)分别是当前系统实际出口压力和质量流量。In formula (17) and formula (18), k is the error correction coefficient, d is the prediction time domain; P_d (k) and m (k) are the actual outlet pressure and mass flow rate of the current system, respectively.10.如权利要求9所述的一种基于预测模型的离心压缩机防喘振控制方法，其特征在于，步骤7中，所述离心压缩机的出口压力和质量流量的性能指标分别如下：10. A kind of predictive model-based centrifugal compressor anti-surge control method as claimed in claim 9, is characterized in that, in step 7, the performance index of the outlet pressure of described centrifugal compressor and mass flow rate are as follows respectively:式(19)和式(20)中，是预测输出误差的加权系数，λj≥0是控制量的加权系数，d是预测时域，Nu控制时域，d≥Nu≥1，v是离心压缩机转速控制输出，k_t是出口节流阀控制输出；In formula (19) and formula (20), is the weighting coefficient of the forecast output error, λj≥0 is the weighting coefficient of the control quantity, d is the forecast time domain, Nu control time domain, d≥Nu≥1, v is the centrifugal compressor speed control output, k_t is the outlet throttling valve control output;所述控制约束条件如下：The control constraints are as follows:s.t. v_min≤v≤v_max (22)，st v_min ≤ v ≤ v_max (22),0≤k_t≤100 (23)，_0≤kt≤100 (23),m_min≤m_p(k+i)≤m_max (24)，m_min ≤ m_p (k+i) ≤ m_max (24),P_d，min≤P_d(k+i)≤P_d，max (25)，P_{d, min} ≤ P_d (k+i) ≤ P_{d, max} (25),P_d，p(k+i)≤P_d，CL(m_p(k+i)) (26)；P_{d, p} (k+i) ≤ P_{d, CL} (m_p (k+i)) (26);式(21)～(26)中，V_min和V_max分别为离心压缩机的最小和最大转速；m_min和m_max分别是离心压缩机最小和最大质量流量，P_d,min和P_d,max分别是离心压缩机最小和最大出口压力，P_d,CL(m_p(k+i))表示在控制线上当质量流量为m_p(k+i)时所对应的出口压力；式(22)与式(23)是执行器本身的约束条件，式(24)和式(25)为离心压缩机工作范围的约束条件，式(26)为防喘振约束条件。In formulas (21) to (26), V_min and V_max are the minimum and maximum rotational speeds of the centrifugal compressor, respectively; m_min and m_max are the minimum and maximum mass flow rates of the centrifugal compressor, respectively, P_d,min and P_{d, max} are the minimum and maximum outlet pressures of the centrifugal compressor respectively, P_d,CL (m_p (k+i)) represents the corresponding outlet pressure when the mass flow rate is m_p (k+i) on the control line; formula (22 ) and formula (23) are the constraints of the actuator itself, formulas (24) and (25) are the constraints of the working range of the centrifugal compressor, and formula (26) is the constraint of anti-surge.