CN102110187B - Method and system for diagnosing mixed failure based on PCA and artificial immune system - Google Patents

Abstract

The invention provides a method and a system for diagnosing mixed failure based on a PCA and an artificial immune system, and the method and the system can be applied in industrial processes as chemical industry, oil refining, biopharmaceutics, and the like. The PCA algorithm is used for detecting failure, the artificial immune system is used for diagnose failure kinds of self and non-self judgments, and self-adaptation and self-study capabilities of the artificial immune system are used, so that the updating of a failure diagnosis system during an online operation process is realized. Antigens and the antibodies of the artificial immune system are represented as a matrix composed of data samples of time series, wherein data to be detected is generated as antigens, all types of historical data are generated as different antibodies, the kind of working condition is judged by computing the difference between the antigen and the antibody, so as to realize failure detection and diagnosis. A clone and variation computation is also presented, so as to produce a large amount of different antibodies from the known working condition data during the failure diagnosis process, and the antibodies are automatically updated during the diagnosis, so that the requirements of actual industrial process on adaptability are met.

Description

A kind of process industry mixed fault diagnostic method and system based on PCA and artificial immune system

Invention field

The present invention relates to a kind of mixed fault diagnostic method and system based on PCA and artificial immune system, it is mainly used in chemical process, oil refining process, the fault diagnosis field of the process industries such as bio-pharmaceuticals process.

Background technology

Along with the development of science and technology, in modern industry was produced, the complexity of commercial unit system improved day by day, for continuous large batch of modern production process, sets up supervisory system, in time finds and diagnose the reason that is out of order, and is very necessary.Fault diagnosis as a comprehensive branch of learning, is the very practical technology that many science are intersected.Effectively, fault diagnosis not only can guarantee the normal operation of producing, the generation of prevention catastrophic failure timely, can also give works engineer to instruct, and in time repairs the position of breaking down, and the effect of directiveness is arranged for production.

Three phases has mainly been experienced in the development of fault diagnosis.First stage is according to sense organ, instrument, relies on operator's experience to judge, this is still more effective for simple and mechanical process.Second stage is to utilize sensor technology and technique of dynamic measurement, and signal is carried out modeling analysis, and this is widely used, and than the previous stage, precision and the standardized degree in this stage are greatly improved.Three phases is just in the developing Intelligent Information Processing stage, and this is also along with technical development of computer, with bionical, the machine-processed a kind of new trial that combines of artificial intelligence.

Fault diagnosis has been passed through development for many years, and theoretical method mainly contains three classes: based on the method for analytic model, Knowledge-Based Method, based on the method for signal, typical method has neural network, least square method, wavelet analysis method, principal component analysis (PCA) etc.Yet traditional method for diagnosing faults does not have large-scale application in the fault diagnosis of process industry, and practical application mainly also is confined to individual equipment, for the fault diagnosis of flow process Shortcomings also itself.Main cause is that the actual industrial flow process has instability, can be along with the improvement of technique and the needs of production change, and traditional method for diagnosing faults scarcely possesses enough adaptability for the variation of flow process, is difficult to satisfy actual process industry needs.For this present situation, the artificial immune system that has the scholar to propose to have adaptivity and self-learning capability is incorporated into fault diagnosis field.

Artificial immune system is a kind of integrated intelligent system, and it organically combines immunology and engineering science, utilizes the technology such as mathematics, computing machine to set up the immunologic mechanism model, and is applied to the aspects such as design, enforcement of engineering.In recent years, in artificial immunity, the judgement for oneself and nonego is incorporated into fault diagnosis field by reference.Yet all variable that system is instantaneous is as antigen and antibody for traditional fault diagnosis algorithm based on artificial immune system, and in the process industry that is beyond expression, data are implemented continuous feature.In view of this, seeking a kind of new antigen and antibody expression method and relevant a series of artificial immune systems thereof is of great significance for the fault diagnosis tool of process industry.

Summary of the invention

The object of the invention is to overcome the defective of existing method applicability and self-learning capability deficiency, artificial immune system is applied to the fault diagnosis of process industry.Be different from traditional artificial immune system, the present invention proposes the expression way of new antigen and antibody and the method for diversity factor calculating and clonal propagation, and proposed a kind of mixed fault diagnostic method based on PCA and artificial immunity.

According to an aspect of the present invention, provide a kind of mixed fault diagnostic method based on PCA and artificial immune system, it is characterized in that comprising:

The PCA modeling procedure generates pca model with original normal sample,

Antibody generates step, generates antibody by the primary fault sample,

Failure detection steps is carried out fault detect with described pca model.

According to a further aspect of the present invention, above-mentioned mixed fault diagnostic method based on PCA and artificial immune system further comprises:

Antigen generates step, in the situation that being determined with fault, described failure detection steps exists, and with the real time data generation antigen of schedule time length before the detection constantly of described failure detection steps,

Troubleshooting step utilizes described antigen and described antibody, judges the kind of fault.

According to a further aspect of the present invention, above-mentioned mixed fault diagnostic method based on PCA and artificial immune system further comprises:

Data acquisition and treatment step gather required flow process variable data, and described flow process variable data comprises historical data and described real time data, and described historical data comprises described original normal sample and described primary fault sample.

According to a further aspect of the present invention, above-mentioned mixed fault diagnostic method based on PCA and artificial immune system further comprises:

Described antigen is carried out clonal vaviation, thereby generates new antibody,

Upgrade antibody in described antibody library with described new antibody,

Determine the described threshold value of described new antibody.

According to another aspect of the present invention, provide a kind of Hybrid fault diagnosis system based on PCA and artificial immune system, it is characterized in that comprising:

The PCA model building device is used for generating pca model with original normal sample,

The antibody generating apparatus is used for generating antibody by the primary fault sample,

Failure detector is used for carrying out fault detect with described pca model.

According to a further aspect of the present invention, above-mentioned Hybrid fault diagnosis system based on PCA and artificial immune system further comprises:

The antigen generating apparatus is used in the situation that described failure detection steps is determined with the fault existence, with the real time data generation antigen of schedule time length before the detection constantly of described failure detection steps,

Trouble-shooter is used for utilizing described antigen and described antibody, judges the kind of fault.

According to a further aspect of the present invention, above-mentioned Hybrid fault diagnosis system based on PCA and artificial immune system further comprises:

Data acquisition and treating apparatus are used for gathering required flow process variable data, and described flow process variable data comprises historical data and described real time data, and described historical data comprises described original normal sample and described primary fault sample.

According to a further aspect of the present invention, above-mentioned Hybrid fault diagnosis system based on PCA and artificial immune system further comprises:

Thereby described antigen is carried out the device that clonal vaviation generates new antibody,

Upgrade the device of the antibody in described antibody library with described new antibody,

Determine the device of the described threshold value of described new antibody.

Description of drawings

Fig. 1 has shown the process flow diagram of method according to an embodiment of the invention.

The formation of the antibody (antigen) that the method according to an embodiment of the invention that shown Fig. 2 adopts.

Antibody (antigen) diversity factor that the method according to an embodiment of the invention that shown Fig. 3 adopts is calculated schematic diagram.

Fig. 4 has shown the schematic diagram when method according to an embodiment of the invention is used to epoxypropane hydration reaction flow process.

Embodiment

The object of the invention is to overcome the defective of existing method applicability and self-learning capability deficiency, artificial immune system is applied to the fault diagnosis of process industry.Be different from traditional artificial immune system, the present invention proposes the expression way of new antigen and antibody and the method for diversity factor calculating and clonal propagation, and proposed a kind of mixed fault diagnostic method based on PCA and artificial immunity.

For the fault diagnosis of process industry, as shown in Figure 1, method according to an embodiment of the invention comprises:

-data acquisition and processing

Hybrid fault diagnosis system according to an embodiment of the invention can directly be connected with production equipment, at step S101, by such as PLC (Programmable Logic Controller), DCS (Distributed Control System (DCS)) and LIMS (Laboratory Information Management System) etc. gather required flow process variable data.Wherein known historical data is used for system initialization, sets up pca model and produces original antibody, and real-time online data is used for fault detection and diagnosis.All data were carried out normalized by formula (1) before initialization or fault diagnosis:

$X=0.5+\frac{x-\stackrel{\‾}{X}}{X_{\max }-X_{\min }}---\left(1\right)$

Wherein x is real data, X, X _maxWith X _minThe mean value of this variable data, maximal value and minimum value when being respectively nominal situation.

according to one embodiment of present invention, data acquisition and treatment step can be used as implements the preposition step that method of the present invention is carried out before.

-pca model is set up

Production data obtains original normal sample after the normalization of step S102, enter the principle modeling that step S103 utilizes PCA (principal component analysis (PCA)).Sample data is denoted as X ∈ R ^{N * m}(N is that specimen sample is counted, and m is the measurand number) calculated the covariance matrix R of X by formula (2).

$R=\frac{1}{n-1}{X}^{T}X---\left(2\right)$

R is carried out svd, and result is suc as formula shown in (3):

$R=\left[P\tilde{P}\right]\mathrm{\Λ}{\left[P\tilde{P}\right]}^T---\left(3\right)$

Λ=diag (λ wherein ₁, λ ₂... λ _m), λ ₁〉=λ ₂〉=... 〉=λ _mBe the eigenwert of R,

Be the matrix that the characteristic of correspondence vector forms, P ∈ R ^{M * k}Be loading matrix, $\tilde{P}\∈R^{m\×(m-k)}$ Loading matrix for residual error.K is the pivot number of model, is determined by formula (4).

$\frac{\underset{i=1}{\overset{k}{\mathrm{\Σ}}}{\mathrm{\λ}}_i}{\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{\mathrm{\λ}}_i}\≤85\%---\left(4\right)$

Obtain respectively pivot score matrix T ∈ R by formula (5) (6) again ^{N * k}Be loading matrix, and the residual error score matrix $\tilde{T}\∈R^{N\×(m-k)}:$

T＝XP (5)

$\tilde{T}=X\tilde{P}---\left(6\right)$

Having obtained like this principal component model is

$X=T{P}^T+\tilde{T}{\tilde{P}}^T---\left(7\right)$

Calculate respectively T by formula (5) (6) again ²The control limit δ of statistic and SPE statistic _TAnd δ _SPE, be used for fault detect

${\mathrm{\δ}}_T=\frac{k(m-1)}{m-k}{F}_{k,m-1,\mathrm{\α}}---\left(8\right)$

${\mathrm{\δ}}_{\mathrm{SPE}}={\mathrm{\θ}}_1{[\frac{C_{\mathrm{\α}}\sqrt{2{\mathrm{\θ}}_2{h}_0^2}}{{\mathrm{\θ}}_1}+1+\frac{{\mathrm{\θ}}_2{h}_0(h_0-1)}{{\mathrm{\θ}}_1^2}]}^{\frac{1}{h_0}}---\left(9\right)$

Wherein α is insolation level, gets the decimal between 0.9 to 1.F _{K, m-1, α}Corresponding insolation level is α, and degree of freedom is k, the F distribution critical value under m-1 sample point. ${\mathrm{\θ}}_i=\underset{j=k+1}{\overset{m}{\mathrm{\Σ}}}{\mathrm{\λ}}_j^i,$ $h_0=1-\frac{2{\mathrm{\θ}}_1{\mathrm{\θ}}_3}{3{\mathrm{\θ}}_2^2},$ C _αCorresponding normal distribution degree of confidence is the statistics fiducial limit of (1-α) %.

So just set up the pca model that can be used for fault detect.

-generation antibody library

For the primary fault sample that obtains after S102 step normalization, at step S104, determine its failure mode and time, get the data that fault is introduced rear set time length, generate first generation antibody, be denoted as Ab_fault=[Ab ₁, Ab ₂..., Ab _n].The matrix that these antibody all are comprised of the seasonal effect in time series data sample, Ab _iBe the time series of each variable, as shown in Figure 2, v-1～v-5 is respectively the data of 5 different variablees of system.First generation antibody of the same type generates a large amount of second generation antibodies through step S105 by formula (10) variation and clone, has consisted of antibody library.

$X^{*}=\left\{\begin{array}{cc}\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{a}_i{X}_i+b(X_c-X_d),& \underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{a}_i=1...n>1\\ (1+\frac{b}{2}){\mathrm{<figure-callout\; id="1"\; label="X"\; filenames="H2009102440664E00022.png,H2009102440664E00031.png"\; state="\{\{state\}\}">X</figure-callout>}}_1& ...n=1\end{array}\right.---\left(10\right)$

Wherein n is the number of known this kind first generation antibody, X _iBe known first generation antibody, X ^*Be the second generation antibody that generates through the variation clone.a _iBe the random decimal in 0 to 1, b is the random decimal between-1 to 1.C and d are the random integers between 1 to n.

Calculate again the threshold value of every kind of fault type antibody.

For two antibody (or antigen) of determining, can adopt the Dynamic Time Warping algorithm to calculate the diversity factor of antigen and antibody.At first get the time series calculated difference degree of identical variable, following d _k(i, j) represents antigen and antibody respectively at i, and j is the Euclidean distance of variable k constantly:

d _k(i，j)＝|Ab _k(j)-Ag _k(i)|j∈[1，n]，i∈[1，n] (13)

Construct a similarity matrix D and be used for describing between antigen and antibody track about the minimum difference degree of this variable.Iterative computation can be by d _k(i, j) obtains D _k(i, j) at first determines the iteration initial point.

D _k(1，1)＝d _k(1，l) (14)

Further iterative computation when j＜n:

$D_k(i,j^{*})=\min \left\{\begin{array}{cc}{D}_k(i-1,j)+d_k(i,j)......l\left(i\right)=j^{*}=j& \mathrm{if\; l}(i-1)\&NotEqual;l(i-2)\\ D_k(i-1,j)+d_k(i,j+1)...l\left(i\right)=j^{*}=j+1& \mathrm{if\; j}<n\\ D_k(i-1,j)+d_k(i,j+2)...l\left(i\right)=j^{*}=j+2& \mathrm{if\; j}<n-1\end{array}\right\}---\left(15\right)$

Stop iteration when j=n, the iterations note of this moment is m=i, has found a shortest calculating path of diversity factor, as shown in Figure 3.Can calculate antigen and this variable diversity factor of antibody is by this iteration result:

${\mathrm{\&eta;}}_k=\frac{\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}|\mathrm{Ab}\left(i\right)-\mathrm{Ag}\left(l\left(i\right)\right)|}{m}---\left(16\right)$

The process stronger to linearity, the diversity factor of same variable are calculated and also can be adopted the straightforward procedure of formula (7) directly to calculate:

${\mathrm{\&eta;}}_k=\frac{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}|\mathrm{Ab}\left(i\right)-\mathrm{Ag}\left(i\right)|}{n}---\left(17\right)$

The diversity factor of calculating all variablees just can obtain the total variances degree vector of antigen and antibody:

η＝[η ₁，η ₂，…，η _n] (18)

Utilize the diversity factor between all antibody of the same type in formula (13)-(18) calculating antibody storehouse, note is η (i, j).The threshold value of being calculated respectively each variable of the type antibody by formula (7) is:

${\mathrm{threshold}}_k=\underset{1\&le;i\&le;\mathrm{<figure-callout\; id="1"\; label="n\; min"\; filenames="H2009102440664E00022.png,H2009102440664E00031.png"\; state="\{\{state\}\}">n</figure-callout>}}{\mathrm{<figure-callout\; id="1"\; label="max"\; filenames="H2009102440664E00022.png,H2009102440664E00031.png"\; state="\{\{state\}\}">max</figure-callout>}}\underset{}{\min }\underset{1\&le;j\&le;n,i\&NotEqual;j}{}\left|{\mathrm{\&eta;}}_k(i,j)\right|---\left(19\right)$

-utilize pca model to carry out fault detect

After step S106 normalization, obtain real-time detection data x _newEnter step S107, utilize PCA to carry out fault detect, comprising:

Calculate respectively T by formula (20) (21) ²Statistic and SPE statistic:

${\mathrm{<figure-callout\; id="2"\; label="T"\; filenames="H2009102440664E00021.png,H2009102440664E00022.png"\; state="\{\{state\}\}">T</figure-callout>}}^2={\left|\right|{\mathrm{\&Lambda;}}_k^{-\frac{1}{2}}T\left|\right|}^2={\left|\right|{\mathrm{\&Lambda;}}_k^{-\frac{1}{2}}{P}^T{x}_{\mathrm{new}}\left|\right|}^2---\left(20\right)$

$\mathrm{SPE}={\left|\right|\tilde{P}{\tilde{P}}^T{x}_{\mathrm{new}}\left|\right|}^2---\left(21\right)$

Wherein

Be-1/2 power of front k diagonal element in Λ, Λ, P,

All when setting up pca model, step S103 obtains;

Compare T ²Whether statistic and SPE statistic, all prescribe a time limit greater than controlling when the statistic such as continuous three points if wherein there is a statistic to think and may break down greater than controlling limit greater than the control limit, enter troubleshooting step, detect next group data otherwise return; And

Alternatively, can also use the statistic of PVR or CVR to carry out fault detect.

-utilize antibody library to carry out fault diagnosis

Break down in case detect confirmation at step S107, process entering step S108, the data generation antigen with set time length before detecting constantly is denoted as Ag=[Ag ₁, Ag ₂..., Ag _n].These antigens are identical with the form of antibody in antibody library, the matrix that all is comprised of the seasonal effect in time series data sample, Ag _iTime series for each variable.

After antigen generates, process the failure mode diagnosis that enters step S109, utilize formula (13)-(18) to calculate the diversity factor of all antibody of antigen and antibody library.When having antibody to calculate diversity factor less than corresponding threshold value, think that fault is consistent with this antibody.If all do not satisfy the threshold value requirement with the diversity factor of all antibody in antibody library, namely be judged as a kind of new fault.

The fault diagnosis result of refer step S109, the operator can carry out Artificial Diagnosis (S201), and is input to system, and system enters the step 110, accepts the result of input as final diagnostic result.

-antibody library upgrades

At step S104, with antigen as first generation antibody.Enter again step S105, by formula (10), the type antibody is carried out clonal vaviation, upgrade the antibody in antibody library, and upgraded the threshold value of this antibody by formula (19).

Embodiment:

The method according to this invention has been applied to the fault diagnosis of the reaction realistic model of an epoxypropane hydration generation propylene glycol.

This process flow diagram flow chart as shown in Figure 4.This reaction is all completed in stirred tank R-1.Pressure in course of reaction in stirred tank is P0, and temperature is T0, and liquid level is L0.R-1 follows two strands and becomes a mandarin, and one is epoxypropane solution, enters R-1 through pump P-1 and valve V-1, wherein contains a small amount of magazine methyl alcohol, and the flow of this stream thigh is F1, and temperature is T1, and propylene oxide concentrations is C1-1, and methanol concentration is C1-2.Another enters R-1 for the reaction water through pump P-2 and valve V-2, and flow is F2, and temperature is T2.

Product enters the T-1 tank through the V-3 valve, and product stream plume amount is F3, and temperature is T3, and product propylene glycol concentration is C3.

Because this reaction is themopositive reaction, need to use the chilled water heat exchange.Chilled water is through V-4 valve and stirred tank heat exchange, and cooling water flow is F4, and temperature is T4.

Also have XC, three controllers of LC and TC in flow process.Wherein, XC is the aperture change flow F1 by valve V-1, to control product propylene glycol concentration C 3.LC is the aperture change flow F3 by valve V-3, to control the liquid level L0 of stirred tank R-1.XC is the aperture change flow F1 by valve V-1, to control product propylene glycol concentration C 3.

The control index of whole process main operating parameters and controller is as shown in table 1.

Table 1 nominal situation parameter arranges

This model mainly comprises five detection variable: the mole fraction-C3 of product propylene glycol, temperature of reaction-T0, product flow-F3, cooling water flow-F4, reactor liquid level-L0.

Carry out fault diagnosis, at first obtain the detection data of these five variablees by step S101, sampling time interval is 5 seconds.

After step S102 is one group of normal data sample normalization of 50000 seconds, then set up pca model at step S103.

In addition, too high raw water temperature T 2, impurity methyl alcohol mole fraction C1-2 is too high and each 2 groups of data of the malfunctioning three kinds of faults of temperature of reaction T0 controller, carry out normalization at step S102,10 sampling numbers according to generating antibody, and generate antibody library variation and the clone that step S105 carries out antibody near step S104 gets the fault introducing time again.

Completed like this initialization of fault diagnosis system, then real time data has been carried out fault detection and diagnosis after step S106 normalization.This example has adopted altogether 13 groups of real time data samples that method is verified.

At first take sample 1 as example, sample 1 is nominal situation, after S106 step normalization, enters S107 step PCA and detects.Calculate the data T of all sampled points ²Statistic and SPE statistic are limit less than controlling always, so testing result thinks that this sample is nominal situation.

Again take sample 2 as example, sample 2 is the too high fault of raw water temperature T 2 that 595s introduces, after S106 step normalization, entering S107 step PCA detects, all limit greater than controlling to T2 statistic and the SPE statistic of continuous three points of 630s at 620s, think fault has occured, enter S108 and go on foot by ten sampling numbers before failure detection time according to generating antigen.Go on foot by S109 again, utilize artificial immune system tracing trouble kind, calculate the diversity factor of all antibody in antigen and antibody library, find in the too high fault antibody library of this antigen and T2 Antibody difference ratio less than the type fault threshold, and with the other types Antibody difference ratio all greater than its corresponding threshold value.Therefore diagnostic result is the too high fault of T2, confirms that through step S110 the Artificial Diagnosis result is T2 too high, enters the S104 step, generates new antibodies by antigen.Go on foot by S105 again, carry out clone and the variation of antibody, upgrade antibody library.

Take sample 11 as example, sample 11 is the fault of the temperature of reaction T0 sensor zero point drift of 290s introducing again, after S106 step normalization, enters S107 step PCA and detects, and arrives the T of continuous three points of 310s at 300s ²Statistic and SPE statistic all greater than controlling limit, are thought fault has occured, and enter S108 and go on foot by ten sampling numbers before failure detection time according to generating antigen.Go on foot by S109 again, utilize artificial immune system tracing trouble kind, calculate the diversity factor of antigen and interior all antibody of antibody library, find this antigen with all Antibody difference ratios all greater than its corresponding threshold value.Therefore diagnostic result thinks that time sample is new fault.Confirm that through step S110 the Artificial Diagnosis result is the drift of T0 sensor zero point.Enter again the S104 step, generate new antibodies by antigen.Go on foot by S105 again, carry out clone and the variation of antibody, upgrade antibody library.

All fault diagnosis results of this example are as shown in table 2.

Table 2 fault diagnosis test result

Sequence number	Fault category	The introducing time	Detection time	Diagnostic-type
					1	Normally	-	-	Normally
2	T2 is too high	595s	630s	T2 is too high
					3	T2 is too high	120s	145s	T2 is too high
4	T2 is too high	230s	255s	T2 is too high
					5	C1-2 is too high	225s	255s	C1-2 is too high
6	C1-2 is too high	315s	340s	C1-2 is too high
					7	C1-2 is too high	320s	350s	C1-2 is too high
8	TC is malfunctioning	240s	350s	TC is malfunctioning

Sequence number	Fault category	The introducing time	Detection time	Diagnostic-type
					9	TC is malfunctioning	205s	230s	TC is malfunctioning
10	TC is malfunctioning	325s	400s	TC is malfunctioning
					11	The drift of T0 sensor zero point	290s	310s	New fault
12	The drift of T0 sensor zero point	465s	485s	The drift of T0 sensor zero point
					13	The drift of T0 sensor zero point	380s	400s	The drift of T0 sensor zero point

Should be understood that, the description of in above narration and explanation, the present invention being carried out just illustrates but not is determinate, and do not breaking away under the prerequisite of the present invention that limits as appended claims, can carry out various changes, distortion and/or revise above-described embodiment.

Claims (4)

1. mixed fault diagnostic method based on PCA and artificial immune system is characterized in that comprising:

The PCA modeling procedure generates pca model with original normal sample,

Antibody generates step, generates antibody by the primary fault sample,

Failure detection steps is carried out fault detect by described pca model,

Antigen generates step, in the situation that being determined with fault, described failure detection steps exists, and with the real time data generation antigen of schedule time length before the detection constantly of described failure detection steps,

Troubleshooting step utilizes described antigen and described antibody, judges the kind of fault,

Data acquisition and treatment step, gather required flow process variable data, described flow process variable data comprises historical data and described real time data, described historical data comprises described original normal sample and described primary fault sample, and wherein said historical data and described real time data are subject to normalized:

$X=0.5+\frac{x-\stackrel{\&OverBar;}{X}}{X_{\max }-X_{\min }}$

Wherein x is real data,

X _maxWith X _minThe mean value of this variable data when being respectively nominal situation, maximal value and minimum value,

Described antibody generates step and comprises:

Determine failure mode and the time of described primary fault sample,

After introducing with fault, the data of the described primary fault sample of schedule time length, generate first generation antibody,

Described first generation antibody is made a variation and clones, thereby generate a large amount of second generation antibodies,

Wherein said first generation antibody and described second generation antibody have consisted of antibody library, and described first generation antibody and described second generation antibody all be used to described troubleshooting step,

Generate second generation antibody by formula (10) variation and clone,

$X*=\left\{\begin{array}{cc}\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{a}_i{X}_i+b(X_c-X_d),& \underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{a}_i=1...n>1\\ (1+\frac{b}{2})X_1& ...n=1\end{array}\right.---\left(10\right)$

Wherein

Ab_fault=[Ab ₁, Ab ₂,, Ab _n] the described first generation antibody of expression,

Ab _iBe the time series of each variable,

N is the number of known this kind first generation antibody,

X _iBe known first generation antibody,

X* is the second generation antibody that generates through the variation clone,

a _iBe the random decimal in 0 to 1,

B is the random decimal between-1 to 1,

C and d are the random integers between 1 to n,

Calculate the threshold value of every kind of fault type antibody,

Described failure detection steps comprises:

Calculate respectively T by formula (20) (21) ²Statistic and SPE statistic:

$T^2={\left|\right|{{\mathrm{\&Lambda;}}_k}^{-\frac{1}{2}}T\left|\right|}^2={\left|\right|{{\mathrm{\&Lambda;}}_k}^{-\frac{1}{2}}{P}^T{x}_{\mathrm{new}}\left|\right|}^2---\left(20\right)$

$\mathrm{SPE}={\left|\right|\tilde{P}{\tilde{P}}^T{x}_{\mathrm{new}}\left|\right|}^2---\left(21\right)$

Wherein

x _newBe described real time data,

Be-1/2 power of front k diagonal element in Λ,

Λ, P, All obtain at described PCA modeling procedure;

Compare T ²Whether statistic and SPE statistic all prescribe a time limit greater than controlling when the statistic of predetermined number greater than controlling limit, and judge and break down,

Described troubleshooting step comprises:

The data of set time length before fault detect are constantly generated antigen A g=[Ag ₁, Ag ₂..., Ag _n], these antigens are identical with the form of antibody in described antibody library, the matrix that all is comprised of the seasonal effect in time series data sample, Ag _iBe the time series of each variable,

Utilize formula (13)-(16) and (18) to calculate the diversity factor of all antibody of antigen and described antibody library:

d _k(i,j)=|Ab _k(j)-Ag _k(i)|j∈[1,n],i∈[1,n](13)

D wherein _k(i, j) represents antigen and antibody respectively at i, and j is the Euclidean distance of variable k constantly,

Construct a similarity matrix D and be used for describing between antigen and antibody track about the minimum difference degree of this variable, iterative computation can be by d _k(i, j) obtains D _k(i, j) at first determines the iteration initial point,

D _k(1,1)=d _k(1,1)(14)

Further iterative computation when j＜n:

$D_k(i,j*)=\min \left\{\begin{array}{ccc}{D}_k(i-1,j)+d_k(i,j)......l\left(i\right)=j*=j& \mathrm{if}& l(i-1)\&NotEqual;l(i-2)\\ D_k(i-1,j)+d_k(i,j+1)...l\left(i\right)=j*=j+1& \mathrm{if}& j<n\\ D_k(i-1,j)+d_k(i,j+2)...l\left(i\right)=j*=j+2& \mathrm{if}& j<n-1\end{array}\right\}---\left(15\right)$

Stop iteration when j=n, the iterations note of this moment is m=i, has found a shortest calculating path of diversity factor,

Can calculate antigen and this variable diversity factor of antibody is by this iteration result:

${\mathrm{\&eta;}}_k=\frac{\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}|\mathrm{Ab}\left(i\right)-\mathrm{Ag}\left(l\left(i\right)\right)|}{m}---\left(16\right)$

The diversity factor of calculating all variablees just can obtain the total variances degree vector of antigen and antibody:

η=[η ₁,η ₂,…,η _n](18)

When having antibody to calculate diversity factor less than corresponding threshold value, judge that fault is consistent with this antibody,

If all do not satisfy the threshold value requirement with the diversity factor of all antibody in described antibody library, namely be judged as a kind of new fault.

2. according to claim 1 the mixed fault diagnostic method based on PCA and artificial immune system is characterized in that further comprising:

By formula (10), described antigen is carried out clonal vaviation, thereby generates new antibody,

Upgrade antibody in described antibody library with described new antibody,

Determine the described threshold value of described new antibody.

3. Hybrid fault diagnosis system based on PCA and artificial immune system is characterized in that comprising:

The PCA model building device is used for generating pca model with original normal sample,

The antibody generating apparatus is used for generating antibody by the primary fault sample,

Failure detector is used for carrying out fault detect with described pca model,

The antigen generating apparatus is used in the situation that described failure detection steps is determined with the fault existence, with the real time data generation antigen of schedule time length before the detection constantly of described failure detection steps,

Trouble-shooter is used for utilizing described antigen and described antibody, judges the kind of fault,

Data acquisition and treating apparatus, be used for gathering required flow process variable data, described flow process variable data comprises historical data and described real time data, described historical data comprises described original normal sample and described primary fault sample, and wherein said historical data and described real time data are subject to normalized:

$X=0.5+\frac{x-\stackrel{\&OverBar;}{X}}{X_{\max }-X_{\min }}$

Wherein x is real data,

X _maxWith X _minThe mean value of this variable data when being respectively nominal situation, maximal value and minimum value,

Described antibody generating apparatus is used for carrying out following operation:

Determine failure mode and the time of described primary fault sample,

After introducing with fault, the data of the described primary fault sample of schedule time length, generate first generation antibody,

Described first generation antibody is made a variation and clones, thereby generate a large amount of second generation antibodies,

Wherein said first generation antibody and described second generation antibody have consisted of antibody library, and described first generation antibody and described second generation antibody all be used to described troubleshooting step,

Generate second generation antibody by following formula (10) variation and clone,

$X*=\left\{\begin{array}{cc}\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{a}_i{X}_i+b(X_c-X_d),& \underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{a}_i=1...n>1\\ (1+\frac{b}{2})X_1& ...n=1\end{array}\right.---\left(10\right)$

Wherein

Ab_fault=[Ab ₁, Ab ₂..., Ab _n] the described first generation antibody of expression,

Ab _iBe the time series of each variable,

N is the number of known this kind first generation antibody,

X _iBe known first generation antibody,

X* is the second generation antibody that generates through the variation clone,

a _iBe the random decimal in 0 to 1,

B is the random decimal between-1 to 1,

C and d are the random integers between 1 to n,

Calculate the threshold value of every kind of fault type antibody,

Described failure detector is used for carrying out following operation:

Calculate respectively T by formula (20) (21) ²Statistic and SPE statistic:

$T^2={\left|\right|{{\mathrm{\&Lambda;}}_k}^{-\frac{1}{2}}T\left|\right|}^2={\left|\right|{{\mathrm{\&Lambda;}}_k}^{-\frac{1}{2}}{P}^T{x}_{\mathrm{new}}\left|\right|}^2---\left(20\right)$

$\mathrm{SPE}={\left|\right|\tilde{P}{\tilde{P}}^T{x}_{\mathrm{new}}\left|\right|}^2---\left(21\right)$

Wherein

x _newBe described real time data,

Be-1/2 power of front k diagonal element in Λ,

Λ, P,

All obtain at described PCA modeling procedure;

Compare T ²Whether statistic and SPE statistic all prescribe a time limit greater than controlling when the statistic of predetermined number greater than controlling limit, and judge and break down,

Described trouble-shooter is used for carrying out following operation:

The data of set time length before fault detect are constantly generated antigen A g=[Ag ₁, Ag ₂..., Ag _n], these antigens are identical with the form of antibody in described antibody library, the matrix that all is comprised of the seasonal effect in time series data sample, Ag _iBe the time series of each variable,

Utilize formula (13)-(16) and (18) to calculate the diversity factor of all antibody of antigen and described antibody library:

d _k(i,j)=|Abx(j)-Ag _k(i)|j∈[1,n],i∈[1,n] (13)

D wherein _k(i, j) represents antigen and antibody respectively at i, and j is the Euclidean distance of variable k constantly,

Construct a similarity matrix D and be used for describing between antigen and antibody track about the minimum difference degree of this variable, iterative computation can be by d _k(i, j) obtains D _k(i, j) at first determines the iteration initial point.

D _k(1,1)=d _k(1,1) (14)

Further iterative computation when j＜n:

$D_k(i,j*)=\min \left\{\begin{array}{ccc}{D}_k(i-1,j)+d_k(i,j)......l\left(i\right)=j*=j& \mathrm{if}& l(i-1)\&NotEqual;l(i-2)\\ D_k(i-1,j)+d_k(i,j+1)...l\left(i\right)=j*=j+1& \mathrm{if}& j<n\\ D_k(i-1,j)+d_k(i,j+2)...l\left(i\right)=j*=j+2& \mathrm{if}& j<n-1\end{array}\right\}---\left(15\right)$

Stop iteration when j=n, the iterations note of this moment is m=i, has found a shortest calculating path of diversity factor,

Can calculate antigen and this variable diversity factor of antibody is by this iteration result:

${\mathrm{\&eta;}}_k=\frac{\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}|\mathrm{Ab}\left(i\right)-\mathrm{Ag}\left(l\left(i\right)\right)|}{m}---\left(16\right)$

The diversity factor of calculating all variablees just can obtain the total variances degree vector of antigen and antibody:

η=[η ₁,η ₂,…,η _n] (18)

When having antibody to calculate diversity factor less than corresponding threshold value, judge that fault is consistent with this antibody,

If all do not satisfy the threshold value requirement with the diversity factor of all antibody in described antibody library, namely be judged as a kind of new fault.

4. according to claim 3 the Hybrid fault diagnosis system based on PCA and artificial immune system, is characterized in that further comprising the antibody updating device, and described antibody updating device further comprises:

By formula (10) thus described antigen is carried out the device that clonal vaviation generates new antibody,

Upgrade the device of the antibody in described antibody library with described new antibody,

Determine the device of the described threshold value of described new antibody.

Images