CN1570627A - Off-line double correction method for assay value based on industrial soft measurement model - Google Patents
Off-line double correction method for assay value based on industrial soft measurement model Download PDFInfo
- Publication number
- CN1570627A CN1570627A CN 200410018393 CN200410018393A CN1570627A CN 1570627 A CN1570627 A CN 1570627A CN 200410018393 CN200410018393 CN 200410018393 CN 200410018393 A CN200410018393 A CN 200410018393A CN 1570627 A CN1570627 A CN 1570627A
- Authority
- CN
- China
- Prior art keywords
- model
- soft
- online
- correction
- laboratory values
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 99
- 238000012937 correction Methods 0.000 title claims description 64
- 238000005259 measurement Methods 0.000 title claims description 33
- 238000003556 assay Methods 0.000 title abstract description 9
- 230000008569 process Effects 0.000 claims abstract description 55
- 238000004519 manufacturing process Methods 0.000 claims abstract description 25
- 238000005096 rolling process Methods 0.000 claims abstract description 11
- 238000005070 sampling Methods 0.000 claims description 33
- 230000009977 dual effect Effects 0.000 claims description 26
- 238000005457 optimization Methods 0.000 claims description 12
- 238000012821 model calculation Methods 0.000 claims description 10
- 238000004422 calculation algorithm Methods 0.000 claims description 8
- 230000008859 change Effects 0.000 claims description 8
- 238000012773 Laboratory assay Methods 0.000 claims description 7
- 238000013016 damping Methods 0.000 claims description 3
- 230000001915 proofreading effect Effects 0.000 claims description 2
- 238000012897 Levenberg–Marquardt algorithm Methods 0.000 claims 1
- 230000006978 adaptation Effects 0.000 claims 1
- 238000012545 processing Methods 0.000 abstract description 6
- 230000008901 benefit Effects 0.000 abstract description 3
- 238000005516 engineering process Methods 0.000 description 17
- 230000000694 effects Effects 0.000 description 14
- 238000004364 calculation method Methods 0.000 description 13
- 230000007774 longterm Effects 0.000 description 10
- 238000004458 analytical method Methods 0.000 description 8
- KKEYFWRCBNTPAC-UHFFFAOYSA-N Terephthalic acid Chemical compound OC(=O)C1=CC=C(C(O)=O)C=C1 KKEYFWRCBNTPAC-UHFFFAOYSA-N 0.000 description 6
- 238000010586 diagram Methods 0.000 description 5
- 239000000126 substance Substances 0.000 description 4
- 230000009471 action Effects 0.000 description 3
- 238000013461 design Methods 0.000 description 3
- 230000007246 mechanism Effects 0.000 description 3
- 230000001590 oxidative effect Effects 0.000 description 3
- 238000009825 accumulation Methods 0.000 description 2
- 238000009776 industrial production Methods 0.000 description 2
- 238000007254 oxidation reaction Methods 0.000 description 2
- 238000011002 quantification Methods 0.000 description 2
- DYNFCHNNOHNJFG-UHFFFAOYSA-N 2-formylbenzoic acid Chemical compound OC(=O)C1=CC=CC=C1C=O DYNFCHNNOHNJFG-UHFFFAOYSA-N 0.000 description 1
- VYPSYNLAJGMNEJ-UHFFFAOYSA-N Silicium dioxide Chemical compound O=[Si]=O VYPSYNLAJGMNEJ-UHFFFAOYSA-N 0.000 description 1
- 239000002253 acid Substances 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 238000004523 catalytic cracking Methods 0.000 description 1
- 238000003889 chemical engineering Methods 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 239000003795 chemical substances by application Substances 0.000 description 1
- 238000004140 cleaning Methods 0.000 description 1
- 230000000052 comparative effect Effects 0.000 description 1
- 238000011217 control strategy Methods 0.000 description 1
- 230000008878 coupling Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 238000009499 grossing Methods 0.000 description 1
- 230000001788 irregular Effects 0.000 description 1
- 239000007788 liquid Substances 0.000 description 1
- 244000005700 microbiome Species 0.000 description 1
- 230000003647 oxidation Effects 0.000 description 1
- 230000000737 periodic effect Effects 0.000 description 1
- 239000003208 petroleum Substances 0.000 description 1
- 238000003908 quality control method Methods 0.000 description 1
- 238000007670 refining Methods 0.000 description 1
- 239000008925 shengjing Substances 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 238000012549 training Methods 0.000 description 1
- 230000001960 triggered effect Effects 0.000 description 1
- 238000009966 trimming Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Images
Landscapes
- Feedback Control In General (AREA)
Abstract
This invention discloses a double correcting method for off-line assay values in the flexible measuring model of industry, which by use of off-line assay values of process indexes to ensure the output of flexible measuring model of high accuracy and tendency. The advantages of this invention are the following: 1, to organically combine the techniques of on-line rolling correcting model and assay values error correcting according to their characteristics; 2, to fully make use of historical error information for weighed processing the current errors to truly reflect the continuity and stability of practical production process; 3, to fully referent the ideas of prediction and control techniques and provide correcting method of on-line model with a theory basis and also to ensure a successful application of the on-ling model correcting techniques in practical process.
Description
Technical field
The present invention relates to a kind of dual bearing calibration of off-line laboratory values based on industrial soft-sensing model.
Background technology
In the process industry field, particularly in the petrochemical industry production run, exist a large amount of being difficult to measure, even the variable that can't measure, generally can only accomplish to rely on the artificial means of chemically examining in laboratory to understand the situation of change of these variablees, this monitors in real time for the realization variable, and technical requirements such as quality control even optimization are far from being enough.So main at present employing is set up soft-sensing model and carried out real-time estimation, promptly the auxiliary variable that is easy to detect by measurement estimates the variable that is difficult to even can't measures based on estimation model.
Because the factors such as imperfection of the interference that generally has of actual industrial production process, time variation, the non-linear and soft-sensing model set up, must consider and solve the on-line correction problem of soft-sensing model, could adapt to actual condition and succeed application.
With the petrochemical complex is in the continuous industry production run of typical case's representative, generally be difficult to even can't set up strict precise math model and describe actual industrial process, that is model error must exist, and the regular variation of possibility, and adopt real-time process model on-line correction technology will improve this problem greatly, thereby can reflect the actual conditions of process comparatively exactly.The kernel model of industrial process soft-measuring technique generally only can roughly reflect the variation tendency of the essence of real process, and realizes that it is very difficult or even impossible that the actual industrial process variable is carried out long-term online accurate estimation.In addition, even can describing process comparatively exactly, model itself changes, but resulting estimation is based on the on-the-spot data message of being gathered according to model, and there is very big uncertainty in the reliability of field data, as the irregular reason such as cleaning of the measuring accuracy drift of the false liquid level of jar, on-line analysis instrument and its, cause data distortion through regular meeting; Simultaneously because generally there are the time lag phenomenon in industrial process itself and relevant parameters measuring process, and retardation time regular fluctuation or the like takes place with the variation of actual industrial production situation, these all may have a strong impact on the reliability of industrial data, these carry out the problem that parameter estimation method ran into based on soft-sensing model and must take into full account, and are exactly effective means and reliable assurance that addresses these problems by means of well behaved on-line correction technology.
Many researchers has all proposed the application need of soft-measuring technique in industrial real process in conjunction with online alignment technique [1,2,3,4,5], but great majority only are the thinkings of simply introducing on-line correction, or have introduced laboratory values to the correction of soft-sensing model output etc., and the on-line correction technology the most close with the art of this patent is the short-term study that proposes of people such as Yu Jinshou [1] and the bearing calibration of long-term study.The short-term correction is to come timely correction model constant term with the deviation of the estimated value of the actual value of certain moment practical object and soft-sensing model, and this alignment technique is easy to canbe used on line.Simultaneously, when deviation is within the error range that technology allows, do not carry out model tuning, have only when deviation to exceed certain limit,, promptly model structure or parameter are revised just excite long-term correction in order to guarantee the stability of soft-sensing model.Short-term study is because algorithm is simple, and pace of learning is fast, is convenient to real-time application.Long-term study is after soft-sensing model on-line operation a period of time has accumulated enough new samples information, rebulids soft-sensing model.Long-term correction can be carried out by off-line, also can onlinely carry out.
Only a constant soft-sensing model that is obtained by off-line is estimated the actual industrial process variable in real time, the uncertain problem of the dynamic change of tracing process, the data reliability in the processing procedure and solve the online variable estimation problem of industrial process of process vary within wide limits well often, therefore, the online soft sensor Model Calculation must be considered to proofread and correct at line process.But alignment technique neither be omnipotent, proofreaies and correct just the process model estimated accuracy to be improved, and soft-sensing model necessarily requires generally to reflect that auxiliary variable in the real process treats the action effect of the leading variable of estimation.
Many researchers has proposed to implement online alignment technique on soft-measuring technique, but simply introduced the laboratory values offset correction mostly or only, perhaps only introduced the technology of carrying out on-line correction, and major part all fails intactly to provide calibration result to based on the network structure and the parameter of neural network soft sensor modeling.People such as Yu Jinshou [1] have proposed the bearing calibration of short-term study and long-term study, but there is not to describe in detail combination how to carry out two kinds of bearing calibrations yet, and at the actual industrial process effect, and only be to introduce each to proofread and correct thinking and independent effect, introduction for two kinds of methods much only is qualitative explanation, to the not enough quantification of some parameter-definitions.The short-term bearing calibration that they propose does not consider to utilize the information of historical deviation, and just utilizes the deviation of current estimated value and actual value to revise, and it is excessive that this method is proved to be meeting generation corrective action fluctuating range in actual applications, and effect is unsatisfactory.Long-term bearing calibration for they propose only relates to after collecting enough sample numbers, and model is proofreaied and correct again, does not further describe in detail and proves or the like.
Implement a kind of alignment technique separately weak point is respectively arranged on improved model estimated accuracy and anticipation trend: the effect that roll modeling is proofreaied and correct is essential, active and violent; The laboratory values offset correction then acts on too passive, and its effect is short-term, can not well proofread and correct the variation of process essence.Therefore the dual alignment technique that organically combines two class alignment techniques can all well be guaranteed on soft-sensing model is from the estimated accuracy to the anticipation trend.This patent has carried out reasonable disposition to two class alignment techniques, obtains optimum correction parameter by experimental verification, can guarantee to make soft-sensing model reach the estimated accuracy of expectation.
List of references
1) Yu Jinshou, Liu Ailun, Zhang Kejin write, soft-measuring technique and the application in petrochemical complex thereof, chemical engineering industry publishing house, 2000.
2) Shi Ruisheng, the characteristics of the online technology Calculation of catalytic cracking, the oil refining design, 2001,31 (1), P.55-59
3) Wang Shuqing etc. writes advanced control technology, Chemical Industry Press, 2000
4) Wang Yongsheng, Shao Huihe, the soft measurement of cell concentration in the microorganism growth process, Wuxi Light Industry Univ.'s journal, 2000,19 (5): 491-494
5) Xu Min, Yu Jinshou, soft-measuring technique, petrochemical complex robotization, 1998,2:1-3,19
6)Shengjing?Mu，Hongye?Su，Ruilan?Liu，Yong?Gu，Jian?Chu，Analysis?andModeling?of?Industrial?Purified?Terephthalic?Acid?Oxidation?Unit，InternationalSymposium?on?Advanced?Control?of?Chemical?Processes，2003，Hongkong.(ADCHEM?2003)
7) Shu Diqian writes, Predictive Control System and application thereof, China Machine Press, 1996
Summary of the invention
The purpose of this invention is to provide a kind of dual bearing calibration of the off-line laboratory values based on industrial soft-sensing model.
It utilizes the off-line laboratory values of in-process metrics, with adjustable cycle the soft-sensing model parameter is carried out roll correction and the dual correction of realization online soft sensor that output valve is carried out offset correction is calculated in soft measurement respectively, make soft-sensing model prediction output have good precision and trend.
Advantage of the present invention:
1) someone has proposed the bearing calibration of short-term study and long-term study, but how not introduce to two kinds of bearing calibrations in conjunction with and effect.And the present invention proofreaies and correct according to online roll modeling and laboratory values offset correction technology characteristics separately organically combine, to give full play to advantage separately;
2) forefathers once separate analysis short-term proofread and correct and long-term simple thinking and the independent effect proofreaied and correct, to the not enough quantification of the introduction of bearing calibration.The short-term bearing calibration that they propose does not consider to utilize the information of historical deviation, and just utilizes the deviation of current predicted value and actual value to revise, and it is excessive that this method is proved to be the corrective action fluctuating range in practice, and effect is unsatisfactory.Laboratory offset correction among the present invention has then made full use of historical deviation information, and current deviation is weighted processing, promptly current deviation has been carried out Filtering Processing, has truly reflected the continuity peace slow fruit of actual production process;
3) the long-term bearing calibration of their proposition is only referred to after collecting enough sample numbers, and model is proofreaied and correct again, does not further describe in detail and demonstration.The online model tuning that the present invention proposes is then being discussed correction mechanism in detail, on the trigger condition basis, thought in the Prediction and Control Technology of in industrial real process, being used widely and two key concepts have wherein been used for reference well: prediction step and modeling step-length, for online model tuning method provides theoretical foundation, this has also guaranteed the application of succeeding of online model tuning technology in real process.
Description of drawings
Fig. 1 is online roll modeling correcting process synoptic diagram;
Fig. 2 is that online roll modeling is proofreaied and correct modeling time domain and prediction time domain synoptic diagram;
Fig. 3 is online laboratory values offset correction schematic flow sheet;
Fig. 4 utilizes laboratory values to carry out online dual correcting process synoptic diagram;
Fig. 5 be soft measurement prediction output (dotted line) and the laboratory values (real light line) of 4-CBA concentration under different bearing calibrations ratio of precision than result schematic diagram;
Fig. 6 be through the soft measurement prediction output of the 4-CBA concentration after the data smoothing (dotted line) and laboratory values (real light line) under different bearing calibrations ratio of precision than result schematic diagram.
Embodiment
The present invention is directed in the industrial process and carry out the situation that the present domestic petrochemical industry process of this class of off-line ground sampling analysis quality variable data the most generally adopts by the laboratory, proofread and correct and laboratory offset correction technology in conjunction with online roll modeling, proposed the online dual alignment technique of online soft sensor model estimation output.Tracking of complex industrial process online soft sensor model prediction trend and estimated accuracy problem can have been solved well through the dual alignment technique of its parameter being done after rationally being provided with.
During the industrial process soft-measuring technique was implemented, soft-sensing model not was unalterable after setting up.Owing to the imperfection of the existing time variation of actual industrial process, non-linear, time lag etc. and the model of setting up based on industrial data and Analysis on Mechanism that obtained and the factors such as reliability of measurement data, the capital influences the precision and the variables corresponding estimated accuracy of soft-sensing model, therefore the on-line correction of model must be considered, actual condition could be adapted to.Present most of online soft sensor model has all been considered on-line correction when design, but seldom has document that this is proposed reasonable and effective solution.
The problem that the soft measurement on-line correction of another one technology must be noted that is that the coupling of data on sequential analyzed in process measurement data and laboratory.For domestic minority petroleum chemical enterprise be equipped with the on-line analysis instrument apparatus and analyze at the scene in obtained the situation of fine application, the assay value of process variable can obtain (lagging behind a period of time) continuously, at timing as long as correspondingly postpone the identical time.But for domestic most of enterprises, its some process variable still relies on the situation of artificial off-line chemical examination, need certain flowing time from the real process variable data to the sample position, expend long time again and return the scene to the experimental analysis data from the back of taking a sample, therefore utilize laboratory values and process data to carry out the soft-sensing model timing, pay particular attention to and keep both corresponding relations in time, otherwise on-line correction not only misses one's aim, may cause the decline of soft measuring accuracy on the contrary, even fall flat.
The present invention utilizes artificial laboratory values that the industrial process soft-sensing model is carried out the dual correction of online soft sensor, promptly combines online roll modeling alignment technique and online laboratory values offset correction technology is implemented effective on-line correction to soft-sensing model.
1 online roll modeling is proofreaied and correct
Online roll modeling is proofreaied and correct, and is the model parameter of proofreading and correct in the soft-sensing model, and the correcting algorithm of utilization is damping nonlinear least square method [6], promptly improved Levenberg-Marquardt (LM) algorithm.The thought of roll correction has been used for reference the rolling optimization and the dynamic modeling thought [7] of Prediction and Control Technology, promptly utilize the up-to-date several groups of sample regression model parameters in long a plurality of sampling periods, the new model that obtains is used to predict ensuing than the variate-value during short several sampling periods.More this just sees and learns several steps, and few walk and do the thought in several steps, the enforcement of rolling, for example, the assay samples data to utilizing 6 points in the soft measurement of carboxyl benzaldehyde (4-CBA) concentration among the thick TA of pure terephthalic acid (PTA) oxidizing process product are carried out model parameter and are returned, qualitative data during the new model that obtains is used to predict follow-up 3 sampling periods, and 3 assay samples that increase newly during utilizing during model tuning next time new relatively in 6 used assay samples of last parametric regression 3 and last model to estimate again are combined into 6 new samples and return sample as up-to-date model parameter and carry out new model tuning once, and the new model that obtains is used to estimate the variate-value during new follow-up 3 sampling periods.The parameter 6 and 3 of bearing calibration has here been used prediction time domain step-length P in the PREDICTIVE CONTROL and control time domain step-length M notion, and its numerical value is adjustable.Modeling time domain that roll modeling is proofreaied and correct and prediction time domain are in time and constantly forward.
The sample that online model tuning is made up of the laboratory values and the corresponding process input variable of online collection some utilizes the nonlinear damping least square method, and promptly improved Levenberg-Marquardt (LM) method is carried out regression iterative to model parameter and calculated.If the regression iterative convergence, model parameter is fetched and is returned result of calculation, if iteration does not restrain, model parameter is still got the model parameter of a calibration cycle.
As shown in Figure 1, online model tuning can regularly be carried out, if the new number of samples m that obtains a sampling period check reaches requirement, then carry out model tuning at once, return new model parameter, then according to the convergence situation of regression algorithm, if do not restrain, model parameter does not change; If a new group model parameter is then got in convergence, obtain new model and continue on-line prediction calculating.
The startup of online roll modeling correcting algorithm is by detecting the accumulation number of laboratory values input sample, carrying out after number reaches the appointment requirement.This is the parameter optimisation procedure of a model, implements by online rolling, dynamically obtains up-to-date process soft-sensing model.In addition, to the model tuning cycle with carry out with resulting up-to-date model in the design in the cycle that on-line prediction calculates, this patent has been used for reference prediction time domain step-length P in the Prediction and Control Technology and control time domain step-length M notion, promptly utilizes sampling instant t
k-P+1 is to t
kBetween the sample data in common P sampling period carry out corrected model parameter and calculate, the model parameter that obtains as up-to-date model to t
k+ 1 to t
kThe process output of+M between the sampling period is predicted.P 〉=M is set here, promptly utilizes the sample information of long time domain to carry out the course prediction output in territory in short-term.After finishing, M sampling period arrive t
k+ M carries out model tuning next time constantly the time, and a required P sample information is to be t by sampling instant
kThe sample of-P+1+M is to t
kThe sample in+M sampling period is combined into P new samples.The modeling time domain of model tuning cycle, model prediction cycle and correspondence and the setting of prediction time domain are as shown in Figure 2 in the online rolling modeling.
At moment t
k, the new samples number reaches and specifies number P, carries out model tuning, utilizes t
kP up-to-date laboratory values sample carries out model parameter and returns constantly, obtain new model after, carry out the output valve in an ensuing M sampling period and predict.Arrive t through after M sampling period
k+ M utilizes one group of new P the sample that comprises up-to-date M laboratory values sample that newly is combined into to carry out model tuning constantly again, obtains new model and carries out the ensuing M prediction of output in a sampling period.Carry out periodic model tuning so online, make new process model the tight tracking process to change, accurately reflect up-to-date process present situation with the adjustment model parameter.
2 laboratory values offset corrections
Because there are a lot of inenarrable factors in actual industrial process, and that the measurable variable of the real process that soft-sensing model can be expressed is treated the influence factor of predictor is also non-very complete, there are other error components in the Model Calculation simultaneously, the result that these reasons make Model Calculation go out can only be an approximate value, only can roughly give expression to the variation tendency of product property, and and often have certain deviation between the assay value of true laboratory, therefore need proofread and correct to eliminate deviation result of calculation.
Online laboratory values offset correction, because the sequence of calculation in the database has only been preserved the estimation output valve of soft-sensing model calculating through overcorrect, and there is not model original calculation value, therefore at first need to obtain the model original calculation value of sampling instant, need try to achieve the model original calculation value of current time equally, obtain the deviation of sampling instant, utilize the deviation and the current time of this sampling instant to calculate the deviation that adopts, the new deviation that current time should adopt is tried to achieve in weighting, thereby further tries to achieve the value that this time correction is crossed.
Detailed step is as follows:
A) obtain the model original calculation value Y of sampling instant
Cal(t-d)
Y
Cal(t-d)＝Y
Val(t-d)-bias(t-d)??????????????????????(1)
Or in the original calculation sequence, find out original calculation value Y
Cal(t-d)
Wherein: d: sampling instant is to the retardation time of current time;
Y
Val(t-d): the soft measurement products quality output valve of gained is proofreaied and correct in sampling instant;
Bias (t-d): the deviation of sampling instant.
B) obtain the model original calculation value of current time
Y
Cal(t)＝Y
Val0(t)-bias(t-d)?????????????????(2)
Wherein: Y
Val0 (t): current time is proofreaied and correct the soft measurement products quality output valve of gained by not corrected deviation (deviate of sampling instant).
C) obtain the deviate bias0 (t) of current time
bias0(t)＝Y
Lab(t-d)-Y
Cal(t-d)???????????????(3)
Wherein: Y
Lab(t-d): the laboratory values that the sampling instant laboratory records (current time just returns).
D) obtain the new deviation bias (t) of current time according to the deviation bias (t-d) of the deviation bias0 (t) of current time and sampling instant
bias(t)＝w
1*bias0(t)+w
2*bias(t-d)???????????(4)
Wherein: w
1=1-w
2, w
1, w
2〉=0 is adjustable correction parameter, generally gets w
1=0.3～0.5.
E) utilize the new deviation of the model original calculation value of current time and current time to obtain the soft measurement products quality output valve Y of current time
Val(t)
Y
Val(t)＝Y
Cal(t)+bias(t)????????????????????(5)
At initial time t=0, bias (0)=0.
Online laboratory values offset correction process flow diagram as shown in Figure 3,
Because the laboratory values input time in current sampling period is late samples moment certain hour always, here comprise two parts time lags, return the on-the-spot spent time to the experimental analysis data from the real process variable data flowing time required with from the back of taking a sample to the sample position.When the user imported the laboratory values in current sampling period, system obtained the model original calculation value Y of sampling instant according to the sampling instant of this laboratory values
Cal(t-d), laboratory values and model original calculation value based on this moment obtain model prediction deviation bias0 (t), deviate with sampling instant is weighted and obtains the current deviate bias (t) constantly that calculates again, utilizes this deviate to proofread and correct current time Model Calculation output Y
Val(t) as last process output.
3 online dual corrections
Online dual alignment technique has organically combined online roll modeling alignment technique and online laboratory values offset correction technology, makes the industrial process soft-sensing model estimate that the precision and the anticipation trend of output can be improved.Model tuning has utilized the laboratory values feedback information to come the adjustment model parameter to change to adapt to up-to-date process better.The laboratory values offset correction then to soft-sensing model can not fine description real process the soft measurement estimated bias problem that causes such as not modeling factor, process data error, laboratory values error done good consideration and solution.Therefore in conjunction with these two kinds of alignment techniques that respectively have superiority, replenish mutually, the dual bearing calibration that obtains can guarantee the estimated accuracy and the anticipation trend of soft-sensing model well.
The industrial process parameter is gathered in real time by on-the-spot DCS, and at moment t, the online Model Calculation input parameter X (t) that reads of soft measuring system is input to Model Calculation and obtains product quality Y
Cal(t).Whether detect moment k has the laboratory values Y of product quality
Lab(t) input if there is not input, then utilizes deviate bias (t) that last offset correction calculates to Model Calculation value Y
Cal(t) proofread and correct Y
Val(t)=Y
Cal(t)+bias (t); If t has the laboratory values input constantly, then accumulate new model tuning number of samples P, and whether the judgement sample number reach setting value M, if the new samples number does not reach the required number of samples of model tuning, then utilize offset correction to carry out new deviation calculation and obtain new bias (t), to Model Calculation value Y
Cal(t) proofread and correct Y
Val(t)=Y
Cal(t)+bias (t); If the new samples number has reached the required number of samples of model tuning, then call the LM algorithm and carry out the model parameter recurrence, finish model tuning, after obtaining new model parameter and setting up new model, call new Model Calculation again and obtain new Y
Cal(t), simultaneously with deviation bias (t) zero clearing, soft measuring system output Y
Val(t)=Y
Cal(t), finish whole trimming process.
Online dual correcting process figure as shown in Figure 4,
Here dual alignment technique combines online roll modeling and proofreaies and correct and the laboratory offset correction, wherein online roll modeling is only proofreaied and correct and is just triggered and enforcement after the new samples number is accumulated to some, and before the new samples number of accumulation does not reach some, laboratory values of every input, the deviation that system-computed is new, and implement the laboratory offset correction.System with the deviate zero clearing, recomputates deviate at every turn when offset correction is carried out in laboratory values input next time after finishing the roll modeling correction.
Bearing calibration of the present invention is analogous to the thought that is adopted in the PREDICTIVE CONTROL strategy, implements from it, has also considered rolling parameter optimization and dynamic line modeling, the deviation feedback compensation that utilizes laboratory values to carry out.Model tuning is to carry out online rolling optimization in time, carries out repeatedly.Wherein each step all is a static optimization, but sees it is dynamic optimization from the overall situation.Every new sampling instant all will be by the actual laboratory values that obtains, i.e. model output is revised the calculating output based on model, and then carries out new optimization.Constantly the prediction output valve is made revising making rolling optimization, and utilized feedback information, constitute closed-loop optimization and proofread and correct not only based on model according to the actual output of system.
4 simulation results relatively
The actual industrial process example application that adopts is the main index of the thick TA character of PTA oxidizing process product of certain PTA factory: the online soft sensor system of 4-CBA content.The process model agent structure is the model structure based on chemical reaction mechanism that is obtained by the laboratory, in conjunction with oxidation reaction process practical operation situation, model is provided with 6 device factors (being the model parameter that needs online rolling to return).Getting the modeling sample number is 6, and predetermined period is 3 laboratory values sampling times, i.e. modeling time domain is 6, and the prediction time domain is 3.Weight w in the offset correction
1=0.5.And in order between laboratory values and soft measurement output valve, to compare, to get soft measurement and calculate output valve and laboratory values and carry out the point of same sampling instant and compare.The field data that is compared has 220 points, and quality index 4-CBA concentration is through normalized.Resulting emulation comparing result as shown in Figure 5.What scheme wherein that a represents is the comparison of calculating output and laboratory assay value through the soft measurement of any treatment for correcting, what figure b represented is that output and the contrast of laboratory assay value are calculated in the soft measurement of only having done the roll modeling treatment for correcting, figure c is that the contrast of exporting with the laboratory assay value is calculated in the soft measurement of only having done the offset correction processing, and figure d is the soft measurement calculating output and the contrast of laboratory assay value of dual treatment for correcting.The statistical error contrast is as shown in table 1.From the comparative result of above-mentioned 4 figure as can be seen, soft-sensing model is done any correction all can significantly improve model estimated accuracy and anticipation trend than not doing correction, and online roll modeling correction is compared with the laboratory offset correction, the former effect is too strong, performance more initiatively, this is to be through the model after the modeling again because online model tuning obtains, what change is model essence estimated performance, show as after system is by new sample training, produce at once the strong effect of subsequent process.And the laboratory offset correction is only revised the calculating of soft-sensing model output, and only works in the soft measurement result of calculation between input of previous laboratory assay value and new next time laboratory values input.The performance of offset correction algorithm is can well follow the tracks of real process to change, this is a passive relatively process, it shows as soft measurement calculating output valve only is to fluctuate round actual laboratory assay value, therefore process is steady more, tracking performance is good more, and when the process amplitude of variation was big, as the lifting load process condition, tracking performance just obviously descended.Therefore the handled estimation output value table of dual bearing calibration that combines two kinds of alignment technique advantageous feature here reveals than two kinds of independent effects that alignment technique is more superior.
The soft measurement precision of prediction of table 1.PTA oxidizing process relatively
| Baseline results | Level and smooth back result | |||
| Mean deviation | Maximum deviation | Mean deviation | Maximum deviation | |
| Without overcorrect | 0.0610 | ?0.2328 | ?0.0501 | ?0.1282 |
| Online roll modeling is proofreaied and correct | 0.0471 | ?0.2614 | ?0.0241 | ?0.0731 |
| The laboratory offset correction | 0.0440 | ?0.2374 | ?0.0176 | ?0.0600 |
| Dual correction | 0.0489 | ?0.2720 | ?0.0142 | ?0.0504 |
For the effect of the dual correction of clearer explanation, by the data point of Fig. 5 is done the sliding time average treatment, do running mean here with 10 points.The intrinsic propesties that the result through simplifying processing who obtains so more can reflect data variation, its result as shown in Figure 6.The effect of dual correction shows as and still all be superior to first three kind situation on the estimated accuracy on anticipation trend.
Claims (5)
1, a kind of to the online dual bearing calibration of the soft measurement of industrial process based on model, it is characterized in that utilizing the off-line laboratory values of in-process metrics, implement the dual bearing calibration that the output valve offset correction is calculated in the roll correction of soft-sensing model parameter and soft measurement respectively with adjustable cycle, make soft-sensing model prediction output have good precision and trend.
2, according to claim 1 a kind of to the online dual bearing calibration of the soft measurement of industrial process based on model, it is characterized in that the said off-line laboratory values of utilizing in-process metrics: be to have utilized the off-line laboratory assay value of calculating pairing process apparatus index with soft-sensing model, be used as soft measuring system is carried out the foundation of on-line correction.
3, according to claim 1 a kind of to the online dual bearing calibration of the soft measurement of industrial process based on model, the roll correction that it is characterized in that said soft-sensing model parameter: be the parameter of proofreading and correct soft-sensing model, the correcting algorithm that is adopted is the damping nonlinear least square method, it is improved Levenberg-Marquardt algorithm, promptly utilize several groups of samples in a plurality of sampling periods to return current model parameter, the variate-value during the model that obtains is used to predict follow-up several sampling period.
4, according to claim 1 a kind of to the online dual bearing calibration of the soft measurement of industrial process based on model, it is characterized in that said soft measurement calculates the output valve offset correction: be to utilize the Model Calculation value of this sampling instant and the deviation between the laboratory values and the soft measurement of current time to calculate the deviation that is adopted, try to achieve the deviation that current time should adopt through weighting, as the offset correction value of current soft measurement calculated value.
5, according to claim 1 a kind of to the online dual bearing calibration of the soft measurement of industrial process based on model, it is characterized in that saidly implementing the dual bearing calibration that a kind of novelty of output valve offset correction is calculated in the roll correction of soft-sensing model parameter and soft measurement respectively with adjustable cycle: be that model tuning has utilized the laboratory values feedback information to adjust the parameter of soft-sensing model, come the procedure of adaptation to change to carry out online rolling optimization in time, laboratory values to the deviation feedback compensation of soft measurement calculated value then be to soft-sensing model the not modeling factor of real process can not be described, the process data error, the soft-sensing model estimated bias that the laboratory values error causes is revised, two class bearing calibrations hocket with the different cycles respectively, this continuous actual output according to system is made revising to the prediction output valve and is made rolling optimization not only based on model, and utilized feedback information, constitute closed-loop optimization and proofread and correct.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN 200410018393 CN1261764C (en) | 2004-05-12 | 2004-05-12 | Off-line double correction method for assay value based on industrial soft measurement model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN 200410018393 CN1261764C (en) | 2004-05-12 | 2004-05-12 | Off-line double correction method for assay value based on industrial soft measurement model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN1570627A true CN1570627A (en) | 2005-01-26 |
| CN1261764C CN1261764C (en) | 2006-06-28 |
Family
ID=34479491
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN 200410018393 Expired - Fee Related CN1261764C (en) | 2004-05-12 | 2004-05-12 | Off-line double correction method for assay value based on industrial soft measurement model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN1261764C (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107544455A (en) * | 2017-08-17 | 2018-01-05 | 浙江邦业科技股份有限公司 | A kind of asymmetric online convergence correction method applied to quality control soft instrument |
| CN109085856A (en) * | 2018-09-10 | 2018-12-25 | 厦门邑通软件科技有限公司 | A kind of desulfuration absorbing tower oxidation fan consumption-reducing method and system |
| WO2019093907A1 (en) | 2017-11-07 | 2019-05-16 | Lic Automation Limited | System and method for analysis of a fluid |
| CN109885869A (en) * | 2019-01-09 | 2019-06-14 | 石化盈科信息技术有限责任公司 | A kind of online customized Technology Calculation |
| CN109933031A (en) * | 2019-03-26 | 2019-06-25 | 沈阳铝镁设计研究院有限公司 | A kind of system and method automatically correcting soft measuring instrument according to analysis data |
| CN110496507A (en) * | 2019-08-12 | 2019-11-26 | 厦门邑通软件科技有限公司 | The method of calcium sulfite concentration is fitted in a kind of wet desulfurizing process |
| CN111758038A (en) * | 2018-02-27 | 2020-10-09 | 三菱电机株式会社 | Method and system for estimating degradation of wire bonded power semiconductor modules |
| CN113405956A (en) * | 2021-06-15 | 2021-09-17 | 中建材（合肥）粉体科技装备有限公司 | On-line correction method, system and equipment for detection data of particle size analyzer |
| CN114791480A (en) * | 2022-03-14 | 2022-07-26 | 国能智深控制技术有限公司 | Soft measurement method and device for dense medium ash content of coal preparation plant |
-
2004
- 2004-05-12 CN CN 200410018393 patent/CN1261764C/en not_active Expired - Fee Related
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107544455B (en) * | 2017-08-17 | 2019-12-31 | 浙江邦业科技股份有限公司 | Asymmetric online convergence correction method applied to quality control soft instrument |
| CN107544455A (en) * | 2017-08-17 | 2018-01-05 | 浙江邦业科技股份有限公司 | A kind of asymmetric online convergence correction method applied to quality control soft instrument |
| WO2019093907A1 (en) | 2017-11-07 | 2019-05-16 | Lic Automation Limited | System and method for analysis of a fluid |
| US11644453B2 (en) | 2017-11-07 | 2023-05-09 | S.C.R. Engineers Limited | System and method for analysis of a fluid |
| EP3707504A4 (en) * | 2017-11-07 | 2021-08-18 | Lic Automation Limited | System and method for analysis of a fluid |
| CN111758038A (en) * | 2018-02-27 | 2020-10-09 | 三菱电机株式会社 | Method and system for estimating degradation of wire bonded power semiconductor modules |
| CN109085856A (en) * | 2018-09-10 | 2018-12-25 | 厦门邑通软件科技有限公司 | A kind of desulfuration absorbing tower oxidation fan consumption-reducing method and system |
| CN109885869B (en) * | 2019-01-09 | 2023-01-20 | 石化盈科信息技术有限责任公司 | Online user-defined process calculation method |
| CN109885869A (en) * | 2019-01-09 | 2019-06-14 | 石化盈科信息技术有限责任公司 | A kind of online customized Technology Calculation |
| CN109933031A (en) * | 2019-03-26 | 2019-06-25 | 沈阳铝镁设计研究院有限公司 | A kind of system and method automatically correcting soft measuring instrument according to analysis data |
| CN109933031B (en) * | 2019-03-26 | 2021-08-31 | 沈阳铝镁设计研究院有限公司 | System and method for automatically correcting soft measuring instrument according to assay data |
| CN110496507B (en) * | 2019-08-12 | 2021-07-13 | 厦门邑通软件科技有限公司 | Method for fitting concentration of calcium sulfite in wet desulphurization process |
| CN110496507A (en) * | 2019-08-12 | 2019-11-26 | 厦门邑通软件科技有限公司 | The method of calcium sulfite concentration is fitted in a kind of wet desulfurizing process |
| CN113405956A (en) * | 2021-06-15 | 2021-09-17 | 中建材（合肥）粉体科技装备有限公司 | On-line correction method, system and equipment for detection data of particle size analyzer |
| CN113405956B (en) * | 2021-06-15 | 2023-07-28 | 中建材（合肥）粉体科技装备有限公司 | On-line correction method, system and equipment for detection data of particle size analyzer |
| CN114791480A (en) * | 2022-03-14 | 2022-07-26 | 国能智深控制技术有限公司 | Soft measurement method and device for dense medium ash content of coal preparation plant |
Also Published As
| Publication number | Publication date |
|---|---|
| CN1261764C (en) | 2006-06-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110378533B (en) | Intelligent aeration management method based on big data analysis | |
| CN1261764C (en) | Off-line double correction method for assay value based on industrial soft measurement model | |
| CN111650834B (en) | Sewage treatment process prediction control method based on extreme learning machine | |
| US11709463B2 (en) | Control method based on adaptive neural network model for dissolved oxygen of aeration system | |
| CN113077039B (en) | Soft measurement method for total nitrogen TN of effluent based on task-driven RBF neural network | |
| CN87106500A (en) | Apparatus and method for using adaptive gain scheduling algorithm | |
| CN108562709A (en) | A kind of sewage disposal system water quality monitoring method for early warning based on convolution self-encoding encoder extreme learning machine | |
| CN109473182B (en) | MBR (Membrane biological reactor) membrane water permeability intelligent detection method based on deep belief network | |
| CN103500280A (en) | Cell concentration online soft measurement method in glutamic acid fermentation process | |
| Imtiaz et al. | Bioreactor profile control by a nonlinear auto regressive moving average neuro and two degree of freedom PID controllers | |
| Urniezius et al. | Generic estimator of biomass concentration for Escherichia coli and Saccharomyces cerevisiae fed-batch cultures based on cumulative oxygen consumption rate | |
| CN116048135B (en) | Photovoltaic cleaning robot endurance optimization method | |
| Li et al. | Investigating the learning effects of technological advancement on CO 2 emissions: A regional analysis in China | |
| CN115305526A (en) | Self-adaptive control method for consistency of copper foil thickness and surface density based on X-ray measurement | |
| CN111142494B (en) | Intelligent control method and system for amine liquid regeneration device | |
| CN114315057B (en) | Water treatment control method, system, computer equipment and storage medium | |
| CN112668234A (en) | Intelligent control method for steelmaking endpoint of converter | |
| CN106054951B (en) | A kind of control method and system of dissolved oxygen concentration | |
| CN107247994B (en) | Fuzzy modeling method for desulfurization efficiency of tray tower desulfurization device | |
| CN1279356C (en) | On-line soft measurement modeling method for 4-CBA content based on least square algorithm of restricted memory part | |
| CN109034366B (en) | Application of ELM integrated model based on multiple activation functions in chemical engineering modeling | |
| CN114519262B (en) | Air target threat prediction method based on improved GM (1, 1) model | |
| CN114315058B (en) | Water treatment control method, system, computer equipment and storage medium | |
| CN115097886A (en) | Method, system, equipment and medium for controlling concentration of dissolved oxygen in sewage treatment | |
| CN1597918A (en) | Process and apparatus for controlling temp. of edible mushroom liquid fermentation |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| C17 | Cessation of patent right | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20060628 Termination date: 20140512 |