CN104133991A - Penicillin fermentation process fault diagnosis method based on kernel partial least squares reconstitution - Google Patents

Abstract

The invention provides a penicillin fermentation process fault diagnosis method based on kernel partial least squares reconstitution. The method comprises the following steps that: off-line historical normal data in the penicillin fermentation process is collected; a penicillin fermentation process operating variable off-line historical normal data set and a penicillin fermentation process state variable off-line historical normal data set are respectively normalized and standardized; an improved kernel partial least squares method is used for building a fault monitoring model of the penicillin fermentation process; faults in the penicillin fermentation process are monitored on line; a penicillin fermentation process fault correlation direction model based on the improved kernel partial least squares reconstitution is built; and the penicillin fermentation process fault diagnosis is carried out. According to the method provided by the invention, an input space is divided into a principal element space directly relevant to the output, a principal element space irrelevant to the output and a residual error space irrelevant to the output. Compared with a traditional method, the penicillin fermentation process fault diagnosis method has the advantages that input variables relevant to the output are monitored, and variables relevant to the input are also precisely monitored.

Description

Based on the penicillin fermentation process method for diagnosing faults of core offset minimum binary reconstruct

Technical field

The invention belongs to Fault monitoring and diagnosis technical field, be specifically related to a kind of penicillin fermentation process method for diagnosing faults based on the reconstruct of core offset minimum binary.

Background technology

Penicillin belongs to antibiotic one, it is a kind of secondary metabolic product, as shown in Figure 1, the whole production cycle comprises 4 physiology phases to its fermentation schematic flow sheet: response lag phase, thalline ramp phase, penicillin synthesis phase and thalline death (self-dissolving) phase; 2 physics sub-periods: cell cultivation stage (is called again the batch operation stage, corresponding the first two physiology phase, approximately continue 45h) and the penicillin fed-batch fermentation stage (be called again the fed-batch operational phase, corresponding latter two physiology phase, approximately continue 355h).In the sweat of penicillin, pH value and temperature adopt closed-loop control, and feed supplement adopts the control of open loop definite value, by the pH value in control course of reaction and the temperature in fermentation reactor, can make reaction carry out under top condition.As secondary microbial metabolism, the common way of this sweat is: carry out first under certain conditions the cultivation of microorganism, this is the initial incubation stage; Then by constantly supplementing glucose, promote the synthetic of penicillin, this is the penicillin fermentation stage, and this stage, required cell all produced in the initial incubation stage.At fermentation stage, penicillin starts to generate as metabolic product, through exponential phase, is continued until repose period.

Because the most penicillin fermentation process automaticities of China are also lower at present, often there will be in process of production the frequent situation about occurring of fault and abnormal conditions.Wherein, controlling when foam, if natural oil used the respiratory metabolism that can affect at most thalline, if the amount of natural oil the speed of impact fermentation at least.In controlling foam, also need to add acid, alkali to regulate the pH value of fermentation, too much or very few acid, alkali all can affect the pH value of fermentation, thereby cause penicillin fermentation failure.In addition, at cell cultivation stage, the concentration of thalline is low, and nutrient culture media is nutritious, easily forms microbiological contamination.Penicillin fermentation in earlier stage, after microbiological contamination, easily breed by miscellaneous bacteria, divides with production bacterium contention nutritional labeling and oxygen, and severe jamming is produced the growth and breeding of bacterium and the generation of product, disturbs the production order, the destruction production schedule; Ferment middle microbiological contamination meeting severe jamming is produced the metabolism of bacterium, affects the generation of product, now, generally can adopt the way of " pouring in down a chimney ", can cause like this increase of a large amount of raw-material wastes and operation cost.Therefore need to diagnose in time the abnormal and fault occurring in penicillin fermentation process.

Summary of the invention

The problem existing for prior art, the invention provides a kind of penicillin fermentation process method for diagnosing faults based on the reconstruct of core offset minimum binary.

Technical scheme of the present invention is achieved in that

Based on the penicillin fermentation process method for diagnosing faults of core offset minimum binary reconstruct, comprise the following steps:

Step 1: gather the historical normal data of off-line of penicillin fermentation process, comprise the historical normal data set of penicillin fermentation process performance variable off-line and the historical normal data set of penicillin fermentation process state variable off-line;

Penicillin fermentation process performance variable comprises the temperature of temperature, pH value and the fermentation reactor of flow velocity, the substrate feeding of ventilation rate, agitator power, substrate feeding;

Penicillin fermentation process state variable comprises the concentration of penicillin, the heat that penicillin reaction produces, concentration, the volume of nutrient culture media and the concentration of thalline of carbon dioxide;

Step 2: respectively the historical normal data set of penicillin fermentation process performance variable off-line and the historical normal data set of penicillin fermentation process state variable off-line are carried out to specification and standardization;

Step 3: utilize improved core offset minimum binary method to set up the malfunction monitoring model of penicillin fermentation process, this model be input as the historical normal data set of penicillin fermentation process performance variable off-line, this model is output as the historical normal data set of penicillin fermentation process state variable off-line;

Step 3.1: data core mapping: historical penicillin fermentation process performance variable off-line normal data set is mapped to high-dimensional feature space by kernel function from original data space;

Step 3.2: utilize kernel partial least squares that historical penicillin fermentation process performance variable off-line normal data set and the historical normal data set of penicillin fermentation process state variable off-line are divided into respectively to principal component space and residual error space;

Step 3.3: utilize principal component space that kernel principal component analysis is residual error space by the residual error spatial division of historical penicillin fermentation process performance variable off-line normal data set and the residual error space in residual error space, and then obtain the malfunction monitoring model of penicillin fermentation process:

Historical penicillin fermentation process performance variable off-line normal data set is expressed as to data set, the data set of principal component space in residual error space and the data set sum in the residual error space in residual error space of its principal component space, historical penicillin fermentation process state variable off-line normal data set is expressed as to the data set of its principal component space and the data set sum in residual error space;

Step 4: utilize the malfunction monitoring model of penicillin fermentation process, the fault of on-line monitoring penicillin fermentation process;

Step 4.1: obtain online penicillin fermentation process performance variable data and penicillin fermentation process state variable data;

Step 4.2: the penicillin fermentation process performance variable data of obtaining online and penicillin fermentation process state variable data are carried out to specification and standardization;

Step 4.3: data core mapping: the penicillin fermentation process performance variable data of obtaining are online mapped to high-dimensional feature space by kernel function from original data space;

Step 4.4: calculate Hotelling statistic in the principal component space of the historical normal data set of penicillin fermentation process performance variable off-line of the penicillin fermentation process performance variable data obtained online, online Hotelling statistic in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line of the penicillin fermentation process performance variable data obtained and the penicillin fermentation process performance variable data obtained the online SPE statistic in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line;

Step 4.5: the Hotelling statistic of the penicillin fermentation process performance variable data of obtaining online that judgement calculates in the principal component space of the historical normal data set of penicillin fermentation process performance variable off-line, in the Hotelling statistic of the penicillin fermentation process performance variable data of obtaining online in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line and the penicillin fermentation process performance variable data obtained the online SPE statistic in the residual error space of penicillin fermentation process performance variable off-line history normal data set, whether there is at least one statistic higher than its corresponding confidence limit: be, current penicillin fermentation process breaks down, execution step 5, otherwise return to step 4.1,

Step 5: set up the relevant direction model of penicillin fermentation process fault based on the reconstruct of improved core offset minimum binary;

Step 5.1: gather the off-line historical failure data of penicillin fermentation process, comprise penicillin fermentation process performance variable off-line historical failure data set and penicillin fermentation process state variable off-line historical failure data set;

Step 5.2: penicillin fermentation process performance variable off-line historical failure data set and penicillin fermentation process state variable off-line historical failure data set are carried out to specification and standardization;

Step 5.3: penicillin fermentation process performance variable off-line historical failure data set is mapped to high-dimensional feature space by kernel function from original data space;

Step 5.4: according to the historical normal data set of penicillin fermentation process performance variable off-line and penicillin fermentation process performance variable off-line historical failure data set, the fault related direction of the principal component space of reconstruct penicillin fermentation process performance variable off-line historical failure data set;

Step 5.4.1: utilize core offset minimum binary method that penicillin fermentation process performance variable off-line historical failure data set and the historical normal data set of penicillin fermentation process performance variable off-line are divided respectively to principal component space and residual error space, obtain respectively the core partial least square model of penicillin fermentation process performance variable off-line historical failure data set and the core partial least square model of the historical normal data set of penicillin fermentation process performance variable off-line;

Step 5.4.2: the data set of the principal component space of historical penicillin fermentation process performance variable off-line normal data set respectively orthogonal mapping to the load matrix in the load matrix of the principal component space of penicillin operational process performance variable off-line historical failure data set and the residual error space of penicillin operational process performance variable off-line historical failure data set, obtain the data set in the residual error space of the principal component space of the penicillin fermentation process performance variable off-line history normal data set after data set and the orthogonal mapping of principal component space of the principal component space of the historical normal data set of penicillin fermentation process performance variable off-line after orthogonal mapping;

Step 5.4.3: utilize kernel principal component analysis to determine the pivot of the data set in the residual error space of the principal component space of the historical normal data set of penicillin fermentation process performance variable off-line after pivot and the orthogonal mapping of data set of the principal component space of the principal component space of the historical normal data set of penicillin fermentation process performance variable off-line after orthogonal mapping;

Step 5.4.4: the pivot in the load matrix direction in the residual error space of the principal component space of the historical normal data set of penicillin fermentation process performance variable off-line of the data set in the pivot in the load matrix direction of the principal component space of the principal component space of the historical normal data set of penicillin fermentation process performance variable off-line of the data set of principal component space of asking respectively penicillin fermentation process performance variable off-line historical failure data set after orthogonal mapping and the residual error space of penicillin fermentation process performance variable off-line historical failure data set after orthogonal mapping, obtains respectively the fault pivot of principal component space of penicillin fermentation process performance variable off-line historical failure data set and the fault pivot in the residual error space of penicillin fermentation process performance variable off-line historical failure data set;

Step 5.4.5: proportion in the pivot of the principal component space of the principal component space of the historical normal data set of penicillin fermentation process performance variable off-line of the fault pivot of the principal component space of calculating penicillin fermentation process performance variable off-line historical failure data set after orthogonal mapping, proportion in the pivot in the residual error space of the principal component space of the historical normal data set of penicillin fermentation process performance variable off-line of the fault pivot in the residual error space of calculating penicillin fermentation process performance variable off-line historical failure data set after orthogonal mapping;

Step 5.4.6: the fault pivot that is greater than the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line after fault pivot and the orthogonal mapping of principal component space of the corresponding penicillin fermentation process performance variable of the each ratio value off-line historical failure data set of the ratio lower limit of setting in two class ratio values forms respectively the pivot direction of principal component space of new penicillin fermentation process performance variable off-line historical failure data set and the pivot direction in the residual error space of new penicillin fermentation process performance variable off-line historical failure data set;

Step 5.4.7: load matrix corresponding to pivot direction of utilizing the residual error space of load matrix corresponding to the pivot direction of principal component space of new penicillin fermentation process performance variable off-line historical failure data set and new penicillin fermentation process performance variable off-line historical failure data set, the load matrix in the residual error space of the principal component space of the load matrix of the principal component space of the principal component space of the historical normal data set of reconstruct penicillin fermentation process performance variable off-line and the historical normal data set of penicillin fermentation process performance variable off-line, it is the fault related direction of the principal component space of penicillin fermentation process performance variable off-line historical failure data set,

Step 5.5: according to the fault related direction of the principal component space in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line and penicillin fermentation process performance variable off-line historical failure Data set reconstruction penicillin fermentation process performance variable off-line historical failure data set;

Step 5.5.1: utilize core pivot element analysis method, be the principal component space in residual error space and the residual error space in residual error space by the residual error spatial division of the residual error space of penicillin fermentation process performance variable off-line historical failure data set and the historical normal data set of penicillin fermentation process performance variable off-line respectively, obtain respectively the kernel pivot model in residual error space of penicillin fermentation process performance variable off-line historical failure data set and the kernel pivot model in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line;

Step 5.5.2: the data set of the principal component space in the residual error space of historical penicillin fermentation process performance variable off-line normal data set respectively orthogonal mapping to the load matrix in the residual error space in the load matrix of the principal component space in the residual error space of penicillin fermentation process performance variable off-line historical failure data set and the residual error space of penicillin fermentation process performance variable off-line historical failure data set, obtain the data set in the residual error space in the residual error space of the penicillin fermentation process performance variable off-line history normal data set after data set and the orthogonal mapping of principal component space in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line after orthogonal mapping;

Step 5.5.3: utilize kernel principal component analysis to determine the pivot of the data set in the residual error space in the residual error space of the penicillin fermentation process performance variable off-line historical failure data set after pivot and the orthogonal mapping of data set of the principal component space in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line after orthogonal mapping;

Step 5.5.4: the pivot in the load matrix direction of the principal component space in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line of the data set of principal component space of asking the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line after orthogonal mapping, pivot in the load matrix direction in the residual error space in the residual error space of the penicillin fermentation process performance variable off-line historical failure data set of the data set in residual error space of asking the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line after orthogonal mapping, it is the fault pivot in the fault pivot of principal component space in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line and the residual error space in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line,

Step 5.5.5: proportion in the pivot of the principal component space in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line of the fault pivot of the principal component space in the residual error space of the historical normal data set of calculating penicillin fermentation process performance variable off-line after orthogonal mapping, proportion in the pivot of fault pivot principal component space in the residual error space of penicillin fermentation process performance variable off-line historical failure data set after orthogonal mapping in the residual error space in the residual error space of the historical normal data set of calculating penicillin fermentation process performance variable off-line;

Step 5.5.6: the fault pivot that is greater than the residual error space in the fault pivot of principal component space in the residual error space of the corresponding penicillin fermentation process performance variable of the each ratio value off-line historical failure data set of the ratio lower limit of setting and the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line in two class ratio values forms respectively the pivot direction in the pivot direction of principal component space in the residual error space of new penicillin fermentation process performance variable off-line historical failure data set and the residual error space in the residual error space of new penicillin fermentation process performance variable off-line historical failure data set;

Step 5.5.7: load matrix corresponding to pivot direction of utilizing the residual error space in the residual error space of load matrix corresponding to the pivot direction of principal component space in the residual error space of new penicillin fermentation process performance variable off-line historical failure data set and new penicillin fermentation process performance variable off-line historical failure data set, the load matrix in the residual error space in the load matrix of the principal component space in the residual error space of reconstruct penicillin fermentation process performance variable off-line historical failure data set and the residual error space of penicillin fermentation process performance variable off-line historical failure data set, it is the fault related direction of the principal component space in the residual error space of penicillin fermentation process performance variable off-line historical failure data set,

Step 5.6: according to the fault related direction in the residual error space in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line and penicillin fermentation process performance variable off-line historical failure Data set reconstruction penicillin fermentation process performance variable off-line historical failure data set;

Step 5.6.1: by the data set in the residual error space in the residual error space of historical penicillin fermentation process performance variable off-line normal data set respectively orthogonal mapping to the load matrix in the residual error space in the load matrix of the principal component space in the residual error space of penicillin fermentation process performance variable off-line historical failure data set and the residual error space of penicillin fermentation process performance variable off-line historical failure data set, obtain the residual error space in the residual error space in the residual error space of the penicillin fermentation process performance variable off-line history normal data set after principal component space and the orthogonal mapping in residual error space in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line after orthogonal mapping;

Step 5.6.2: utilize kernel principal component analysis to determine the pivot of the data set in the residual error space in the residual error space in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line after pivot and the orthogonal mapping of data set of the principal component space in the residual error space in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line after orthogonal mapping;

Step 5.6.3: the load matrix that the data set of the principal component space in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line is mapped to the principal component space in the residual error space in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line after orthogonal mapping, the data set in the residual error space in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line is mapped to the load matrix in the residual error space in the residual error space in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line after orthogonal mapping, obtain the data set of the mapping space in the residual error space in the data set of mapping space of the principal component space in the residual error space in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line and the residual error space in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line,

Step 5.6.4: the data set of the principal component space in the residual error space of calculating penicillin fermentation process performance variable off-line historical failure data set is to the distance of the load matrix of the data set of the mapping space of the principal component space in the residual error space in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line, the data set in the residual error space in the residual error space of the historical normal data set of calculating penicillin fermentation process performance variable off-line is to the distance of the load matrix of the data set of the mapping space in the residual error space in the residual error space in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line, and calculate the ratio of two distances,

Step 5.6.5: the data set in the residual error space in the residual error space of calculating penicillin fermentation process performance variable off-line historical failure data set is to the distance of the load matrix of the data set of the mapping space of the principal component space in the residual error space in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line, the data set in the residual error space in the residual error space of the historical normal data set of calculating penicillin fermentation process performance variable off-line is to the distance of the load matrix of the data set of the mapping space in the residual error space in the residual error space in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line, and calculate the ratio of two distances,

Step 5.6.6: the ratio of this two classes distance is greater than the pivot direction that the principal component space in residual error space of penicillin fermentation process performance variable off-line historical failure data set that the ratio of its all distances of setting ratio lower limit is corresponding and the residual error space in the residual error space of penicillin fermentation process performance variable off-line historical failure data set form respectively the residual error space in the pivot direction of principal component space in the residual error space in the residual error space of new penicillin fermentation process performance variable off-line historical failure data set and the residual error space in the residual error space of new penicillin fermentation process performance variable off-line historical failure data set;

Step 5.6.7: utilize the pivot direction in the residual error space in the pivot direction of principal component space in the residual error space in the residual error space of new penicillin fermentation process performance variable off-line historical failure data set and the residual error space in the residual error space of new penicillin fermentation process performance variable off-line historical failure data set, the fault related direction in the residual error space in the residual error space of the historical normal data set of reconstruct penicillin fermentation process performance variable off-line;

Step 5.7: for all faults of penicillin fermentation process, applying step 5.1-5.6 reconstructs respectively the fault related direction of the principal component space of penicillin fermentation process performance variable off-line historical failure data set, the fault related direction of the principal component space in the residual error space of penicillin fermentation process performance variable off-line historical failure data set, the fault related direction in the residual error space in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line forms the penicillin fermentation process fault related direction based on the reconstruct of improved core offset minimum binary, thereby set up the model bank of the penicillin fermentation process fault related direction of improved core offset minimum binary reconstruct,

Step 6: utilize the relevant direction model of penicillin fermentation process fault to carry out penicillin fermentation process fault diagnosis;

Step 6.1: according to the fault related direction of the principal component space of penicillin fermentation process performance variable off-line historical failure data set, calculate score matrix and the corresponding Hotelling statistic thereof of the principal component space of the penicillin fermentation process performance variable data of obtaining online;

Step 6.2: according to the fault related direction of the principal component space in the residual error space of penicillin fermentation process performance variable off-line historical failure data set, calculate score matrix and the corresponding Hotelling statistic thereof in the residual error space of the penicillin fermentation process performance variable data of obtaining online;

Step 6.3: according to the fault related direction in the residual error space in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line, calculate the SPE statistic in the residual error space of the penicillin fermentation process performance variable data of obtaining online;

Step 6.4: the Hotelling statistic of the principal component space of the penicillin fermentation process performance variable data obtained online of judgement, online the residual error space of the penicillin fermentation process performance variable data obtained Hotelling statistic, whether the SPE statistic in the residual error space of the penicillin fermentation process performance variable data obtained all lower than the confidence limit separately of setting online, be,, when prior fault is fault real in penicillin fermentation process, be out of order combination or new fault otherwise work as prior fault.

Beneficial effect:

1. in the monitoring of traditional kernel partial least squares: have some variablees and output orthogonal in (1) principal component space, these variablees are irrelevant with output; (2) kernel partial least squares is not as kernel principal component analysis, the method of successively decreasing according to eigenwert is extracted pivot, therefore the variable in residual error space is not applied the problem of SPE statistic monitoring for classic method, the present invention is divided into the input space: with the directly related principal component space of output, with the irrelevant principal component space (principal component space in residual error space) of output and with the irrelevant residual error space (the residual error space in residual error space) of output.Compared with classic method, both monitored the input variable relevant with output, monitor accurately again and input relevant variable.

2,, thus the method for improved Kernel partial least squares regression is reconstructed to the different faults type monitoring in penicillin fermentation process.By the method for improved Kernel partial least squares regression, the performance variable that production run is obtained has carried out correlativity processing, use data after treatment to be reconstructed and to set up improved core offset minimum binary reconstruct monitoring model, by the simulation experiment result, validity of the present invention and feasibility have been described.

Brief description of the drawings

Fig. 1 is penicillin fermentation schematic flow sheet;

Fig. 2 is the malfunction monitoring statistical graph of fault 1 data of the specific embodiment of the invention;

(a) be the Hotelling statistics spirogram that obtains online the principal component space of penicillin fermentation process performance variable data;

(b) be the Hotelling statistics spirogram that obtains online the residual error space of penicillin fermentation process performance variable data;

(c) for obtaining online the SPE in the residual error space of penicillin fermentation process performance variable data _xstatistics spirogram;

Wherein, curve 1: the Hotelling statistic of obtaining online the principal component space of penicillin fermentation process performance variable data; Curve 2: the confidence limit of obtaining online the Hotelling statistic of the principal component space of penicillin fermentation process performance variable data; Curve 3: the Hotelling statistic of obtaining online the residual error space of penicillin fermentation process performance variable data; Curve 4: the confidence limit of obtaining online the Hotelling statistic in the residual error space of penicillin fermentation process performance variable data; Curve 5: the SPE statistic SPE that obtains online the residual error space of penicillin fermentation process performance variable data _x; Curve 6: the SPE statistic SPE that obtains online the residual error space of penicillin fermentation process performance variable data _xconfidence limit;

Fig. 3 is the malfunction monitoring statistical graph of fault 2 data of the specific embodiment of the present invention;

(a) be the Hotelling statistics spirogram that obtains online the principal component space of penicillin fermentation process performance variable data;

(b) be the Hotelling statistics spirogram that obtains online the residual error space of penicillin fermentation process performance variable data;

(c) for obtaining online the SPE in the residual error space of penicillin fermentation process performance variable data _xstatistics spirogram;

Wherein, curve 7: the Hotelling statistic of obtaining online the principal component space of penicillin fermentation process performance variable data; Curve 8: the confidence limit of obtaining online the Hotelling statistic of the principal component space of penicillin fermentation process performance variable data; Curve 9: the Hotelling statistic of obtaining online the residual error space of penicillin fermentation process performance variable data; Curve 10: the confidence limit of obtaining online the Hotelling statistic in the residual error space of penicillin fermentation process performance variable data; Curve 11: the SPE statistic SPE that obtains online the residual error space of penicillin fermentation process performance variable data _x; Curve 12: the SPE statistic SPE that obtains online the residual error space of penicillin fermentation process performance variable data _xconfidence limit;

Fig. 4 is the fault diagnosis statistics spirogram of fault 1 data of the specific embodiment of the present invention;

(a) be the Hotelling statistics spirogram of the principal component space of the performance variable data of fault related direction tracing trouble 1 data of fault 1 data foundation;

(b) be the Hotelling statistics spirogram in the residual error space of the performance variable data of fault related direction tracing trouble 1 data of fault 1 data foundation;

(c) be the SPE statistic SPE in the residual error space of the performance variable data of fault related direction tracing trouble 1 data of fault 1 data foundation _xfigure;

Wherein, curve 13: the Hotelling statistic of the principal component space of the performance variable data of fault related direction tracing trouble 1 data that fault 1 data are set up; Curve 14: the confidence limit of the Hotelling statistic of the principal component space of the performance variable data of fault related direction tracing trouble 1 data that fault 1 data are set up; Curve 15: the Hotelling statistic in the residual error space of the performance variable data of fault related direction tracing trouble 1 data that fault 1 data are set up; Curve 16: the confidence limit of the Hotelling statistic in the residual error space of the performance variable data of fault related direction tracing trouble 1 data that fault 1 data are set up; Curve 17: the SPE statistic SPE in the residual error space of the performance variable data of fault related direction tracing trouble 1 data that fault 1 data are set up _x; Curve 18: the SPE statistic SPE in the residual error space of the performance variable data of fault related direction tracing trouble 1 data that fault 1 data are set up _xconfidence limit;

Fig. 5 is the fault diagnosis statistics spirogram of fault 1 data of the specific embodiment of the present invention;

(a) be the Hotelling statistics spirogram of the principal component space of the performance variable data of fault related direction tracing trouble 1 data of fault 2 data foundation;

(b) be the Hotelling statistics spirogram in the residual error space of the performance variable data of fault related direction tracing trouble 1 data of fault 2 data foundation;

(c) be the SPE statistic SPE in the residual error space of the performance variable data of fault related direction tracing trouble 1 data of fault 2 data foundation _xfigure;

Wherein, curve 19: the Hotelling statistic of principal component space of the performance variable data of fault related direction tracing trouble 1 data of setting up for fault 2 data; Curve 20: the confidence limit of Hotelling statistic of the principal component space of the performance variable data of fault related direction tracing trouble 1 data of setting up for fault 2 data; Curve 21: the Hotelling statistic in residual error space of the performance variable data of fault related direction tracing trouble 1 data of setting up for fault 2 data; Curve 22: the confidence limit of Hotelling statistic in the residual error space of the performance variable data of fault related direction tracing trouble 1 data of setting up for fault 2 data; Curve 23: the SPE statistic SPE in residual error space of the performance variable data of fault related direction tracing trouble 1 data of setting up for fault 2 data _x; Curve 24: the SPE statistic SPE in residual error space of the performance variable data of fault related direction tracing trouble 1 data of setting up for fault 2 data _xconfidence limit;

Fig. 6 is the fault diagnosis statistics spirogram of fault 2 data of the specific embodiment of the present invention;

(a) be the Hotelling statistics spirogram of the principal component space of the performance variable data of fault related direction tracing trouble 2 data of fault 2 data foundation;

(b) be the Hotelling statistics spirogram in the residual error space of the performance variable data of fault related direction tracing trouble 2 data of fault 2 data foundation;

(c) be the SPE statistic SPE in the residual error space of the performance variable data of fault related direction tracing trouble 2 data of fault 2 data foundation _xfigure;

Wherein, curve 25: the Hotelling statistic of principal component space of the performance variable data of fault related direction tracing trouble 2 data of setting up for fault 2 data; Curve 26: the confidence limit of Hotelling statistic of the principal component space of the performance variable data of fault related direction tracing trouble 2 data of setting up for fault 2 data; Curve 27: the Hotelling statistic in residual error space of the performance variable data of fault related direction tracing trouble 2 data of setting up for fault 2 data; Curve 28: the confidence limit of Hotelling statistic in the residual error space of the performance variable data of fault related direction tracing trouble 2 data of setting up for fault 2 data; Curve 29: the SPE statistic SPE in residual error space of the performance variable data of fault related direction tracing trouble 2 data of setting up for fault 2 data _x; Curve 30: the SPE statistic SPE in residual error space of the performance variable data of fault related direction tracing trouble 2 data of setting up for fault 2 data _xconfidence limit;

Fig. 7 is the fault diagnosis statistics spirogram of fault 2 data of the specific embodiment of the present invention;

(a) be the Hotelling statistics spirogram of the principal component space of the performance variable data of fault related direction tracing trouble 2 data of fault 1 data foundation;

(b) be the Hotelling statistics spirogram in the residual error space of the performance variable data of fault related direction tracing trouble 2 data of fault 1 data foundation;

(c) be the SPE statistic SPE in the residual error space of the performance variable data of fault related direction tracing trouble 2 data of fault 1 data foundation _xfigure;

Wherein, curve 31: the Hotelling statistic of principal component space of the performance variable data of fault related direction tracing trouble 2 data of setting up for fault 1 data; Curve 32: the confidence limit of Hotelling statistic of the principal component space of the performance variable data of fault related direction tracing trouble 2 data of setting up for fault 1 data; Curve 33: the Hotelling statistic in residual error space of the performance variable data of fault related direction tracing trouble 2 data of setting up for fault 1 data; Curve 34: the confidence limit of Hotelling statistic in the residual error space of the performance variable data of fault related direction tracing trouble 2 data of setting up for fault 1 data; Curve 35: the SPE statistic SPE in residual error space of the performance variable data of fault related direction tracing trouble 2 data of setting up for fault 1 data _x; Curve 36: the SPE statistic SPE in residual error space of the performance variable data of fault related direction tracing trouble 2 data of setting up for fault 1 data _xconfidence limit;

Fig. 8 is the penicillin fermentation process method for diagnosing faults process flow diagram based on the reconstruct of core offset minimum binary of the specific embodiment of the present invention.

Embodiment

Below in conjunction with accompanying drawing, the specific embodiment of the present invention is elaborated.

Because the fault in penicillin fermentation process is Protean, even if the input of the different faults of same performance variable is also different on the impact of state variable, but finally all can affect output and the quality of penicillin.

In present embodiment there is two kinds of dissimilar faults---Class1 and type 2 in the ventilation rate of penicillin fermentation process, represent respectively step fault and slope fault, the penicillin fermentation process method for diagnosing faults of application based on the reconstruct of core offset minimum binary diagnosed for the fault type 1 of ventilation rate and the penicillin fermentation process of 2 two kinds of different fault types of type.

Based on the penicillin fermentation process method for diagnosing faults of core offset minimum binary reconstruct, as shown in Figure 8, comprise the following steps:

Step 1: gather the historical normal data of off-line of penicillin fermentation process, comprise the historical normal data set Χ '=[x of penicillin fermentation process performance variable off-line ₁..., x _b] ∈ (J × B) and the historical normal data set Y '=[y of penicillin fermentation process state variable off-line ₁..., y _b] ∈ (J ' × B), wherein, the number that subscript B is data sampling, J ', J are variable number;

Penicillin fermentation process performance variable comprises the temperature of temperature, pH value and the fermentation reactor of flow velocity, the substrate feeding of ventilation rate, agitator power, substrate feeding;

Penicillin fermentation process state variable comprises the concentration of penicillin, the heat that penicillin reaction produces, concentration, the volume of nutrient culture media and the concentration of thalline of carbon dioxide;

The historical normal data set of each penicillin fermentation process performance variable off-line and the historical normal data set of penicillin fermentation process state variable off-line respectively comprise 300 samples;

Step 2: respectively to the historical normal data set Χ '=[x of penicillin fermentation process performance variable off-line ₁..., x _b] and the historical normal data set Y '=[y of penicillin fermentation process state variable off-line ₁..., y _b] carry out specification and standardization: make respectively the historical normal data set Χ '=[x of penicillin fermentation process performance variable off-line ₁..., x _b] and the historical normal data set Y '=[y of penicillin fermentation process state variable off-line ₁..., y _b] average be 0 and variance be 1, obtain the historical normal data set Χ=[x of specification and standardized penicillin fermentation process state variable off-line ₁..., x _b] and specification and the historical normal data set Y=[y of standardized penicillin fermentation process state variable off-line ₁..., y _b];

Step 3: utilize improved core offset minimum binary method to set up the malfunction monitoring model of penicillin fermentation process, this model be input as the historical normal data set Χ=[x of penicillin fermentation process performance variable off-line ₁..., x _b], this model is output as the historical normal data set Y=[y of penicillin fermentation process state variable off-line ₁..., y _b];

Step 3.1: data core mapping: historical penicillin fermentation process performance variable off-line normal data set is mapped to high-dimensional feature space by kernel function Φ from original data space, i.e. X → Φ (X);

Step 3.2: utilize kernel partial least squares that historical penicillin fermentation process performance variable off-line normal data set Χ and the historical normal data set Y of penicillin fermentation process state variable off-line are divided into respectively to principal component space and residual error space;

Step 3.3: utilize principal component space that kernel principal component analysis is residual error space by the residual error spatial division of historical penicillin fermentation process performance variable off-line normal data set Χ and the residual error space in residual error space, and then obtain the malfunction monitoring model of penicillin fermentation process: the data set that historical penicillin fermentation process performance variable off-line normal data set is expressed as to its principal component space, the data set sum in the residual error space in the data set of the principal component space in residual error space and residual error space, historical penicillin fermentation process state variable off-line normal data set is expressed as to the data set of its principal component space and the data set sum in residual error space,

$\left\{\begin{array}{c}\mathrm{\Φ}{\left(X\right)}^T={\mathrm{TP}}^T+T_x{P_x}^T+\mathrm{\Φ}{\left(\stackrel{\‾}{X}\right)}^T\\ Y^T=T{Q}^T+{\stackrel{\‾}{Y}}^T\end{array}\right.---\left(1\right)$

Wherein, subscript T represents transposition, TP ^tfor the data set of the principal component space of the historical normal data set Χ of penicillin fermentation process performance variable off-line, for the data set in the residual error space of the historical normal data set Χ of penicillin fermentation process performance variable off-line, T _xp _x ^tfor the data set of the principal component space in the residual error space of the historical normal data set Χ of penicillin fermentation process performance variable off-line, for the data set in the residual error space in the residual error space of the historical normal data set Χ of penicillin fermentation process performance variable off-line, TQ ^tfor the data set of the principal component space of the historical normal data set Y of penicillin fermentation process state variable off-line, for the data set in the residual error space of the historical normal data set Y of penicillin fermentation process state variable off-line; T is the score matrix of the principal component space of the historical normal data set Χ of penicillin fermentation process performance variable off-line, and P is the load matrix of the principal component space of the historical normal data set Χ of penicillin fermentation process performance variable off-line, T _xfor the score matrix of the principal component space in the residual error space of the historical normal data set Χ of penicillin fermentation process performance variable off-line, P _xfor the load matrix of the principal component space in the residual error space of the historical normal data set Χ of penicillin fermentation process performance variable off-line;

Step 4: utilize the malfunction monitoring model of penicillin fermentation process, the fault of on-line monitoring penicillin fermentation process;

Step 4.1: obtain online penicillin fermentation process performance variable data x ' _new∈ (J × 1) and penicillin fermentation process state variable data y ' _new∈ (J ' × 1);

Step 4.2: the penicillin fermentation process performance variable data of obtaining online and penicillin fermentation process state variable data are carried out to specification and standardization, obtain the penicillin fermentation process performance variable data x after specification and standardization _newpenicillin fermentation process state variable data y after ∈ (J × 1) and specification and standardization _new∈ (J ' × 1);

Step 4.3: data core mapping: the penicillin fermentation process performance variable data of obtaining are online mapped to high-dimensional feature space by kernel function Φ from original data space, i.e. x _new→ Φ (x _new);

Step 4.4: calculate the penicillin fermentation process performance variable data x obtaining online _newhotelling statistic (Hotelling-T in the principal component space of the historical normal data set Χ of penicillin fermentation process performance variable off-line ²), the penicillin fermentation process performance variable data x that obtains online _newhotelling statistic (Hotelling-T in the residual error space of the historical normal data set Χ of penicillin fermentation process performance variable off-line ²) and the penicillin fermentation process performance variable data x that obtains online _newat the historical normal data set x of penicillin fermentation process performance variable off-line _newresidual error space in SPE statistic;

The penicillin fermentation process performance variable data x obtaining online _newhotelling statistic in the principal component space of the historical normal data set Χ of penicillin fermentation process performance variable off-line is as follows:

$\left\{\begin{array}{c}{t}_{\mathrm{new}}=\mathrm{\Φ}\left(x_{\mathrm{new}}\right)\mathrm{\Φ}{\left(X\right)}^T{u}_k=K_t{u}_k\\ T^2=t_{\mathrm{new}}{\mathrm{\Λ}}^{-1}{t}_{\mathrm{new}}^T\end{array}\right.---\left(2\right)$

In formula, t _newfor the penicillin fermentation process performance variable data x obtaining online _newnonlinear principal component in high-dimensional feature space; K _tfor the penicillin fermentation process performance variable data x obtaining online _newwith the inner product vector of the historical normal data set Χ of penicillin fermentation process performance variable off-line, u _kfor the latent variable of the historical normal data set Χ of penicillin fermentation process performance variable off-line; Λ is the covariance matrix of the historical normal data set Χ of penicillin fermentation process performance variable off-line; T ²represent the penicillin fermentation process performance variable data x obtaining online _newhotelling statistic (Hotelling-T in the principal component space of the historical normal data set Χ of penicillin fermentation process performance variable off-line ²);

The penicillin fermentation process performance variable data x obtaining online _newscore vector and Hotelling statistic in the residual error space of the historical normal data set Χ of penicillin fermentation process performance variable off-line are as follows:

$\left\{\begin{array}{c}{t}_{x,\mathrm{new}}={P_x}^T\mathrm{\Φ}\left({\hat{x}}_{\mathrm{new}}\right)\\ T_x^2=t_{x,\mathrm{new}}{\mathrm{\Λ}}_x^{-1}{t}_{x,\mathrm{new}}^T\end{array}\right.---\left(3\right)$

In formula, t _{x, new}for the penicillin fermentation process performance variable data x obtaining online _newresidual error nonlinear principal component in high-dimensional feature space; for x from original data space is mapped to high-dimensional feature space _newresidual error, Λ _xit is the covariance matrix in the residual error space of the historical normal data set Χ of penicillin operational process off-line; represent x _newhotelling statistic in the residual error space of the historical normal data set Χ of penicillin operational process off-line;

The penicillin fermentation process performance variable data x obtaining online _newsPE statistic in the residual error space of the historical normal data set Χ of penicillin operational process off-line is as follows:

$\left\{\begin{array}{c}{e}_{x,\mathrm{new}}=(I-P_x{P_x}^T)\mathrm{\Φ}\left({\hat{x}}_{\mathrm{new}}\right)\\ {\mathrm{SPE}}_x={\left|\right|e_{x,\mathrm{new}}\left|\right|}^2\end{array}\right.---\left(4\right)$

In formula, e _{x, new}for the penicillin fermentation process performance variable data x obtaining online _newresidual error residual error in high-dimensional feature space, SPE _xfor the penicillin fermentation process performance variable data x obtaining online _newsPE statistic in the residual error space of the historical normal data set Χ of penicillin operational process off-line;

Step 4.5: the penicillin fermentation process performance variable data x obtaining online that judgement calculates _newhotelling statistic T in the principal component space of the historical normal data set Χ of penicillin fermentation process performance variable off-line ², the penicillin fermentation process performance variable data x that obtains online _newhotelling statistic in the residual error space of the historical normal data set Χ of penicillin fermentation process performance variable off-line the penicillin fermentation process performance variable data x obtaining online _newsPE statistic SPE in the residual error space of the historical normal data set Χ of penicillin fermentation process performance variable off-line _xin whether have at least one statistic higher than its corresponding confidence limit: be, current penicillin fermentation process breaks down, execution step 5, otherwise return to step 4.1;

Step 5: set up the relevant direction model of penicillin fermentation process fault based on the reconstruct of improved core offset minimum binary;

Step 5.1: gather the off-line historical failure data of penicillin fermentation process, comprise penicillin fermentation process performance variable off-line historical failure data set Χ ' _f=[x _1f..., x _bf] ∈ (J × B) and penicillin fermentation process state variable off-line historical failure data set Y _f'=[y _1f..., y _bf] ∈ (J ' × B);

Step 5.2: to penicillin fermentation process performance variable off-line historical failure data set Χ ' _f=[x _1f..., x _bf] and penicillin fermentation process state variable off-line historical failure data set Y _f'=[y _1f..., y _bf] carry out specification and standardization:

Make respectively penicillin fermentation process performance variable off-line historical failure data set Χ ' _f=[x _1f..., x _bf] and penicillin fermentation process state variable off-line historical failure data set Y _f'=[y _1f..., y _bf] average be 0 and variance be 1, obtain specification and standardized penicillin fermentation process performance variable off-line historical failure data set Χ _f=[x _1f..., x _bf] and specification and standardized penicillin fermentation process state variable off-line historical failure data set Y _f=[y _1f..., y _bf];

Step 5.3: penicillin fermentation process performance variable off-line historical failure data set is mapped to high-dimensional feature space by kernel function Φ from original data space, i.e. X _f→ Φ (X _f);

Step 5.4: according to the historical normal data set Χ of penicillin fermentation process performance variable off-line and penicillin fermentation process performance variable off-line historical failure data set Χ _f, reconstruct penicillin fermentation process performance variable off-line historical failure data set Χ _fthe fault related direction of principal component space;

Step 5.4.1: utilize core offset minimum binary method by penicillin fermentation process performance variable off-line historical failure data set Χ _fhistorical normal data set Χ divides respectively principal component space and residual error space with penicillin fermentation process performance variable off-line, obtains respectively penicillin fermentation process performance variable off-line historical failure data set Χ _fcore partial least square model Φ (X _f) ^tcore partial least square model Φ (X) with the historical normal data set Χ of penicillin fermentation process performance variable off-line ^t;

$\begin{array}{c}\mathrm{\Φ}{\left(X\right)}^T=\mathrm{\Φ}{\left(\tilde{X}\right)}^T+\mathrm{\Φ}{\left(\hat{X}\right)}^T=T{P}^T+\mathrm{\Φ}{\left(\hat{X}\right)}^T=\mathrm{\Φ}{\left(X\right)}^T{\mathrm{PP}}^T+\mathrm{\Φ}{\left(\hat{X}\right)}^T\\ =\mathrm{\Φ}{\left(X\right)}^T\mathrm{\Φ}\left(X\right){\mathrm{\α\α}}^T\mathrm{\Φ}{\left(X\right)}^T+\mathrm{\Φ}{\left(\hat{X}\right)}^T=K{\mathrm{\α\α}}^T\mathrm{\Φ}{\left(X\right)}^T+\mathrm{\Φ}{\left(\hat{X}\right)}^T\end{array}---\left(5\right)$

$\begin{array}{c}\mathrm{\Φ}{\left(X_f\right)}^T=\mathrm{\Φ}{\left({\tilde{X}}_f\right)}^T+\mathrm{\Φ}{\left({\hat{X}}_f\right)}^T=T_f{P_f}^T+\mathrm{\Φ}{\left(X_f\right)}^T{P}_f^{*}{P}_f^{{*}^T}=\mathrm{\Φ}{\left(X_f\right)}^T{P}_f{P}_f^T+\mathrm{\Φ}{\left(X_f\right)}^T{P}_f^{*}{P}_f^{{*}^T}\\ =\mathrm{\Φ}{\left(X_f\right)}^T\mathrm{\Φ}\left(X_f\right){\mathrm{\α}}_f{{\mathrm{\α}}_f}^T\mathrm{\Φ}{\left(X_f\right)}^T+\mathrm{\Φ}{\left(X_f\right)}^T{P}_f^{*}{P}_f^{{*}^T}=K_f{\mathrm{\α}}_f{{\mathrm{\α}}_f}^T\mathrm{\Φ}{\left(X_f\right)}^T+\mathrm{\Φ}{\left(X_f\right)}^T{P}_f^{*}{P}_f^{{*}^T}\end{array}---\left(6\right)$

In formula, for the data set of the principal component space of the historical normal data set Χ of penicillin fermentation process performance variable off-line, for the data set in the residual error space of the historical normal data set Χ of penicillin fermentation process performance variable off-line, for penicillin fermentation process performance variable off-line historical failure data set Χ _fthe data set of principal component space, for penicillin fermentation process performance variable off-line historical failure data set Χ _fthe data set in residual error space, α is the proper vector of the principal component space of Φ (X), α _ffor Φ (X _f) the proper vector of principal component space, P _ffor penicillin fermentation process performance variable off-line historical failure data set Χ _fthe load matrix of principal component space, for penicillin fermentation process performance variable off-line historical failure data set Χ _fthe load matrix in residual error space, K is the historical normal data set Χ of penicillin fermentation process performance variable off-line and the interior product matrix of penicillin fermentation process performance variable off-line history normal data set Χ, K _ffor penicillin fermentation process performance variable off-line historical failure data set Χ _fwith penicillin fermentation process performance variable off-line historical failure data set Χ _finterior product matrix, and P=Φ (X) α, K=Φ (X) ^tΦ (X), P _f=Φ (X _f) α _f, K _f=Φ (X _f) ^tΦ (X _f);

Step 5.4.2: the data set of the principal component space of the historical normal data set Χ of penicillin fermentation process performance variable off-line orthogonal mapping is to penicillin operational process performance variable off-line historical failure data set Χ respectively _fthe load matrix P of principal component space _fwith penicillin operational process performance variable off-line historical failure data set Χ _fthe load matrix P in residual error space _f ^*, obtain the books of the principal component space of the principal component space of the historical normal data set Χ of penicillin fermentation process performance variable off-line after orthogonal mapping data set with the residual error space of the principal component space of the historical normal data set Χ of penicillin fermentation process performance variable off-line after orthogonal mapping

$\begin{array}{c}\mathrm{\Φ}{\left({\tilde{X}}_{\mathrm{nf}}\right)}^T=\mathrm{\Φ}{\left(\tilde{X}\right)}^T{P}_f{P_f}^T=\mathrm{\Φ}{\left(X\right)}^{T}P{P}^T{P}_f{P_f}^T\\ =\mathrm{\Φ}{\left(X\right)}^T\mathrm{\Φ}\left(X\right){\mathrm{\α\α}}^T\mathrm{\Φ}{\stackrel{}{\left(X\right)}}^T{\mathrm{\Φ}}_f\left(X\right){\mathrm{\α}}_f{{\mathrm{\α}}_f}^T\mathrm{\Φ}{\left(X_f\right)}^T\\ =K{\mathrm{\α\α}}^T{K}_N{\mathrm{\α}}_f{{\mathrm{\α}}_f}^T\mathrm{\Φ}{\left(X_f\right)}^T\end{array}---\left(7\right)$

$\begin{array}{c}\mathrm{\Φ}{\left({\tilde{X}}_{\mathrm{nf}}\right)}^{{*}^T}=\mathrm{\Φ}{\left(\tilde{X}\right)}^T{P_f}^{*}{P}_f^{{*}^T}=\mathrm{\Φ}{\left(X\right)}^{T}P{P}^T{P_f}^{*}{P_f}^{{*}^T}\\ =\mathrm{\Φ}{\left(X\right)}^T\mathrm{\Φ}\left(X\right){\mathrm{\α\α}}^T\mathrm{\Φ}{\stackrel{}{\left(X\right)}}^T{\mathrm{\Φ}}_f\left(X\right){{\mathrm{\α}}_f}^{*}{{\mathrm{\α}}_f}^{{*}^T}\mathrm{\Φ}{\left(X_f\right)}^T\\ =K{\mathrm{\α\α}}^T{K}_N{{\mathrm{\α}}_f}^{*}{{\mathrm{\α}}_f}^{{*}^T}\mathrm{\Φ}{\left(X_f\right)}^T\end{array}---\left(8\right)$

In formula, for Φ (X _f) the proper vector in residual error space, Κ _nfor the historical normal data set Χ of penicillin fermentation process performance variable off-line and penicillin operational process performance variable off-line historical failure data set Χ _finterior product matrix, k _n=Φ (X) ^tΦ (X _f);

Step 5.4.3: utilize kernel principal component analysis to determine the principal component space data set of the principal component space of the historical normal data set Χ of penicillin fermentation process performance variable off-line after orthogonal mapping pivot T _cresidual error space data sets with the principal component space of the historical normal data set Χ of penicillin fermentation process performance variable off-line after orthogonal mapping pivot

$\begin{array}{c}{T}_c=\mathrm{\Φ}{\left({\tilde{X}}_{\mathrm{nf}}\right)}^T\\ =K{\mathrm{\α\α}}^T{K}_N{{\mathrm{\α}}_f{\mathrm{\α}}_f}^T\mathrm{\Φ}{\left(X_f\right)}^T\mathrm{\Φ}\left(X_f\right){\mathrm{\α}}_f{{\mathrm{\α}}_f}^T{K_N}^T{\mathrm{\α\α}}^T{K}^T{\mathrm{\α}}_c\\ =K{\mathrm{\α\α}}^T{K}_N{\mathrm{\α}}_f{{\mathrm{\α}}_f}^T{K}_f{\mathrm{\α}}_f{{\mathrm{\α}}_f}^T{K_N}^T{\mathrm{\α\α}}^T{K}^T{\mathrm{\α}}_c\end{array}---\left(9\right)$

$\begin{array}{c}{T_c}^{*}=\mathrm{\Φ}{\left({\tilde{X}}_{\mathrm{nf}}\right)}^{{*}^T}{p_c}^{*}\\ =K{\mathrm{\α\α}}^T{K}_N{{\mathrm{\α}}_f}^{*}{{\mathrm{\α}}_f}^{{*}^T}{K}_f{{\mathrm{\α}}_f}^{*}{{\mathrm{\α}}_f}^{{*}^T}{K_N}^T{\mathrm{\α\α}}^T{K}^T{{\mathrm{\α}}_c}^{*}\end{array}---\left(10\right)$

In above formula, α _cfor the proper vector of principal component space, for the proper vector of principal component space, P _cfor load matrix, for load matrix, $P_c=\mathrm{\Φ}\left({\tilde{X}}_{\mathrm{nf}}\right){\mathrm{\α}}_c,P_c^{*}=\mathrm{\Φ}{\left({\tilde{X}}_{\mathrm{nf}}\right)}^{*}{\mathrm{\α}}_c^{*};$

Step 5.4.4: ask respectively penicillin fermentation process performance variable off-line historical failure data set Χ _fthe data set of principal component space the load matrix P of the principal component space of the principal component space of the historical normal data set Χ of penicillin fermentation process performance variable off-line after orthogonal mapping _cpivot in direction and penicillin fermentation process performance variable off-line historical failure data set Χ _fthe data set in residual error space the load matrix in the residual error space of the principal component space of the historical normal data set of penicillin fermentation process performance variable off-line after orthogonal mapping pivot in direction, obtains respectively the fault pivot T of the principal component space of penicillin fermentation process performance variable off-line historical failure data set _fcfault pivot T with the residual error space of penicillin fermentation process performance variable off-line historical failure data set _fc ^*;

$\begin{array}{c}{T}_{\mathrm{fc}}\text{=\Φ}{\left({\tilde{X}}_f\right)}^T{P}_c=\mathrm{\Φ}{\left({\tilde{X}}_f\right)}^T\mathrm{\Φ}\left({\tilde{X}}_{\mathrm{nf}}\right){\mathrm{\α}}_c\\ =\mathrm{\Φ}{\left(X_f\right)}^T{P}_f{P_f}^T\mathrm{\Φ}\left({\tilde{X}}_{\mathrm{nf}}\right){\mathrm{\α}}_c\\ =K_f{\mathrm{\α}}_f{{\mathrm{\α}}_f}^T{K}_N{\mathrm{\α}}_f{{\mathrm{\α}}_f}^T{K_N}^T{\mathrm{\α\α}}^T{K}^T{\mathrm{\α}}_c\end{array}---\left(11\right)$

$\begin{array}{c}{T_{\mathrm{fc}}}^{*}\text{=\Φ}{\left({\hat{X}}_f\right)}^T{P}_c^{*}=\mathrm{\Φ}{\left({\hat{X}}_f\right)}^T\mathrm{\Φ}{\left({\tilde{X}}_{\mathrm{nf}}\right)}^{*}{{\mathrm{\α}}_c}^{*}\\ =\mathrm{\Φ}{\left(X_f\right)}^T{P_f}^{*}{P_f}^{{*}^T}\mathrm{\Φ}{\left({\tilde{X}}_{\mathrm{nf}}\right)}^{*}{{\mathrm{\α}}_c}^{*}\\ =K_f{{\mathrm{\α}}_f}^{*}{{\mathrm{\α}}_f}^{{*}^T}{K}_N{{\mathrm{\α}}_f}^{*}{{\mathrm{\α}}_f}^{{*}^T}{K_N}^T{\mathrm{\α\α}}^T{K}^T{{\mathrm{\α}}_c}^{*}\end{array}---\left(12\right)$

Step 5.4.5: the fault pivot T that calculates the principal component space of penicillin fermentation process performance variable off-line historical failure data set _fcthe pivot T of the principal component space of the principal component space of the historical normal data set of penicillin fermentation process performance variable off-line after orthogonal mapping _cmiddle proportion RT _{fc, i}, calculate the fault pivot T in the residual error space of penicillin fermentation process performance variable off-line historical failure data set _fc ^*the pivot T in the residual error space of the principal component space of the historical normal data set of penicillin fermentation process performance variable off-line after orthogonal mapping _c ^*middle proportion

${\mathrm{RT}}_{\mathrm{fc},i}=\frac{\mathrm{var}\left(T_{\mathrm{fc}}(:,i)\right)}{\mathrm{var}\left(T_c(:,i)\right)}(i=\mathrm{1,2},\·\·\·,R_c)---\left(13\right)$

${\mathrm{RT}}_{\mathrm{fc},i}^{*}=\frac{\mathrm{var}\left(T_{\mathrm{fc}}^{*}(:,i)\right)}{\mathrm{var}\left(T_c^{*}(:,i)\right)}(i=\mathrm{1,2},\·\·\·,R_c^{*})---\left(14\right)$

In formula, var () represents variance, (:, i) represent that the i of a matrix is listed as, the fault pivot T in above formula _fcat pivot T _cmiddle proportion RT _{fc, i}with fault pivot T _fc ^*at pivot T _c ^*middle proportion larger, represent fault data principal component space larger to the contribution of Hotelling statistic along each load direction (being the direction of the load vector representative in load matrix);

Step 5.4.6: the fault pivot that is greater than the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line after fault pivot and the orthogonal mapping of principal component space of the corresponding penicillin fermentation process performance variable of the each ratio value off-line historical failure data set of the ratio lower limit of setting in two class ratio values forms respectively the pivot direction R of the principal component space of new penicillin fermentation process performance variable off-line historical failure data set _fcpivot direction with the residual error space of new penicillin fermentation process performance variable off-line historical failure data set

Step 5.4.7: the pivot direction R that utilizes the principal component space of new penicillin fermentation process performance variable off-line historical failure data set _fccorresponding load matrix P _fcpivot direction with the residual error space of new penicillin fermentation process performance variable off-line historical failure data set corresponding load matrix P _fc ^*, the load matrix P of the principal component space of the principal component space of the historical normal data set Χ of reconstruct penicillin fermentation process performance variable off-line _cload matrix P with the residual error space of the principal component space of the historical normal data set of penicillin fermentation process performance variable off-line _c ^*, i.e. the fault related direction P of the principal component space of penicillin fermentation process performance variable off-line historical failure data set _{rec, c};

$P_{\mathrm{rec},c}=[P_{\mathrm{fc}},P_{\mathrm{fc}}^{*}]=[\mathrm{\Φ}\left({\tilde{X}}_{\mathrm{nf}}\right){\mathrm{\α}}_{\mathrm{fc}},\mathrm{\Φ}{\left({\tilde{X}}_{\mathrm{nf}}\right)}^{*}{{\mathrm{\α}}_{\mathrm{fc}}}^{*}]---\left(15\right)$

Wherein, α _fcfor with pivot direction R _fccharacteristic of correspondence vector, α _fc ^*for with pivot direction characteristic of correspondence vector;

Step 5.5: according to the historical normal data set Χ of penicillin fermentation process performance variable off-line and penicillin fermentation process performance variable off-line historical failure data set Χ _freconstruct penicillin fermentation process performance variable off-line historical failure data set Χ _fthe fault related direction of principal component space in residual error space;

Step 5.5.1: utilize core pivot element analysis method, respectively by penicillin fermentation process performance variable off-line historical failure data set Χ _fresidual error space and the residual error spatial division of the historical normal data set Χ of penicillin fermentation process performance variable off-line be the principal component space in residual error space and the residual error space in residual error space, obtain respectively the kernel pivot model in the residual error space of penicillin fermentation process performance variable off-line historical failure data set kernel pivot model with the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line

In formula, for the data set of the principal component space in the residual error space of the historical normal data set Χ of penicillin fermentation process performance variable off-line, for the data set in the residual error space in the residual error space of the historical normal data set Χ of penicillin fermentation process performance variable off-line, for penicillin fermentation process performance variable off-line historical failure data set Χ _fthe data set of principal component space in residual error space, E _f ^tfor penicillin fermentation process performance variable off-line historical failure data set Χ _fthe data set in residual error space in residual error space, P _xfor the load matrix in the residual error space of the historical normal data set Χ of penicillin fermentation process performance variable off-line, P _x ^*for penicillin fermentation process performance variable off-line historical failure data set Χ _fthe load matrix in residual error space, α _ifor the proper vector of principal component space, α _i ^*for the proper vector in residual error space, α _iffor the proper vector of principal component space, α _if ^*for the proper vector in residual error space, P _x,ffor penicillin fermentation process performance variable off-line historical failure data set Χ _fthe load matrix of principal component space in residual error space, P _x,f ^*for penicillin fermentation process performance variable off-line historical failure data set Χ _fthe load matrix in residual error space in residual error space, K _ifor the data set in the residual error space of the historical normal data set Χ of penicillin fermentation process performance variable off-line data set with the residual error space of the historical normal data set Χ of penicillin fermentation process performance variable off-line interior product matrix, K _iffor penicillin fermentation process performance variable off-line historical failure data set Χ _fthe data set in residual error space with penicillin fermentation process performance variable off-line historical failure data set Χ _fthe data set in residual error space interior product matrix, and $P_x=\mathrm{\Φ}\left(\hat{X}\right){\mathrm{\α}}_i,{P_x}^{*}=\mathrm{\Φ}\left(\hat{X}\right){{\mathrm{\α}}_i}^{*},K_i=\mathrm{\Φ}{\left(\hat{X}\right)}^T\mathrm{\Φ}\left(\hat{X}\right),P_{x,f}=\mathrm{\Φ}\left({\hat{X}}_f\right){\mathrm{\α}}_{\mathrm{if}},{P_{x,f}}^{*}=\mathrm{\Φ}\left({\hat{X}}_f\right){{\mathrm{\α}}_{\mathrm{if}}}^{*},$ $K_{\mathrm{if}}=\mathrm{\Φ}{\left({\hat{X}}_f\right)}^T\mathrm{\Φ}\left({\hat{X}}_f\right);$

Step 5.5.2: the data set of the principal component space in the residual error space of the historical normal data set Χ of penicillin fermentation process performance variable off-line orthogonal mapping is to penicillin fermentation process performance variable off-line historical failure data set Χ respectively _fthe load matrix P of principal component space in residual error space _x,fwith penicillin fermentation process performance variable off-line historical failure data set Χ _fthe load matrix P in residual error space in residual error space _x,f ^*, obtain the data set of the principal component space in the residual error space of the historical normal data set Χ of penicillin fermentation process performance variable off-line after orthogonal mapping data set with the residual error space in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line after orthogonal mapping

In formula, Κ _mfor the data set in the residual error space of the historical normal data set Χ of penicillin fermentation process performance variable off-line with penicillin fermentation process performance variable off-line historical failure data set Χ _fresidual error space data sets interior product matrix, and $K_M=\mathrm{\Φ}{\left(\hat{X}\right)}^T\mathrm{\Φ}\left({\hat{X}}_f\right);$

Step 5.5.3: utilize kernel principal component analysis to determine the data set of the principal component space in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line after orthogonal mapping pivot T _pdata set with the residual error space in the residual error space of the penicillin fermentation process performance variable off-line historical failure data set after orthogonal mapping pivot T _p ^*;

In formula, α _pfor the proper vector of principal component space, for the proper vector of principal component space, P _pfor load matrix, for load matrix,

Step 5.5.4: the data set of asking the principal component space in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line the load matrix P of the principal component space in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line after orthogonal mapping _ppivot in direction, asks the data set in the residual error space in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line the load matrix in the residual error space in the residual error space of the penicillin fermentation process performance variable off-line historical failure data set after orthogonal mapping pivot in direction, i.e. the fault pivot T of the principal component space in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line _fpfault pivot T with the residual error space in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line _fp ^*;

Step 5.5.5: the fault pivot T that calculates the principal component space in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line _fpthe pivot T of the principal component space in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line after orthogonal mapping _pmiddle proportion, calculates the fault pivot in the residual error space in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line the pivot of the principal component space in the residual error space of penicillin fermentation process performance variable off-line historical failure data set after orthogonal mapping middle proportion;

${\mathrm{RT}}_{\mathrm{fp},i}=\frac{\mathrm{var}\left(T_{\mathrm{fp}}(:,i)\right)}{\mathrm{var}\left(T_p(:,i)\right)}(i=\mathrm{1,2},\·\·\·,R_p)---\left(24\right)$

${\mathrm{RT}}_{\mathrm{fp},i}^{*}=\frac{\mathrm{var}\left(T_{\mathrm{fp}}^{*}(:,i)\right)}{\mathrm{var}\left(T_p^{*}(:,i)\right)}(i=\mathrm{1,2},\·\·\·,R_p^{*})---\left(25\right)$

In above formula, fault pivot T _fpat pivot T _pmiddle proportion RT _{fp, i}with fault pivot in pivot middle proportion larger, the residual error space of expression penicillin fermentation process performance variable off-line historical failure data set along each load direction to Hotelling statistic contribute larger;

Step 5.5.6: the fault pivot that is greater than the fault pivot of principal component space in the residual error space of the corresponding penicillin fermentation process performance variable of the each ratio value off-line historical failure data set of the ratio lower limit of setting and the residual error space in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line in two class ratio values forms respectively the pivot direction R of the principal component space in the residual error space of new penicillin fermentation process performance variable off-line historical failure data set _fppivot direction with the residual error space in the residual error space of new penicillin fermentation process performance variable off-line historical failure data set

Step 5.5.7: the pivot direction R that utilizes the principal component space in the residual error space of new penicillin fermentation process performance variable off-line historical failure data set _fpcorresponding load matrix P _fppivot direction with the residual error space in the residual error space of new penicillin fermentation process performance variable off-line historical failure data set corresponding load matrix the load matrix P of the principal component space in the residual error space of reconstruct penicillin fermentation process performance variable off-line historical failure data set _pload matrix with the residual error space in the residual error space of penicillin fermentation process performance variable off-line historical failure data set be the fault related direction P of the principal component space in the residual error space of penicillin fermentation process performance variable off-line historical failure data set _{rec, p};

Wherein, α _fpfor with R _fpcharacteristic of correspondence vector, α _fp ^*for with characteristic of correspondence vector;

Step 5.6: according to the historical normal data set Χ of penicillin fermentation process performance variable off-line and penicillin fermentation process performance variable off-line historical failure data set Χ _freconstruct penicillin fermentation process performance variable off-line historical failure data set Χ _fthe fault related direction in residual error space in residual error space;

Step 5.6.1: by the data set E in the residual error space in the residual error space of the historical normal data set Χ of penicillin fermentation process performance variable off-line ^torthogonal mapping is to penicillin fermentation process performance variable off-line historical failure data set Χ respectively _fthe load matrix P of principal component space in residual error space _x,fload matrix P with the residual error space in the residual error space of penicillin fermentation process performance variable off-line historical failure data set _x,f ^*, obtain the principal component space in the residual error space in the residual error space of the historical normal data set Χ of penicillin fermentation process performance variable off-line after orthogonal mapping residual error space with the residual error space in the residual error space of the historical normal data set Χ of penicillin fermentation process performance variable off-line after orthogonal mapping

$\begin{array}{c}{E}_{\mathrm{nf}}^T=E^T{P}_{x,f}{P}_{x,f}^T=\mathrm{\Φ}{\left(\hat{X}\right)}^T{P_x^{*}{P_x^{*}}^{T}P}_{x,f}{P}_{x,f}^T\\ =K_i{\mathrm{\α}}_i^{*}{\mathrm{\α}}_i^{{*}^T}\mathrm{\Φ}{\left(\hat{X}\right)}^T\mathrm{\Φ}\left({\hat{X}}_f\right){\mathrm{\α}}_{\mathrm{if}}{\mathrm{\α}}_{\mathrm{if}}^T\mathrm{\Φ}{\left({\hat{X}}_f\right)}^T\\ =K_i{\mathrm{\α}}_i^{*}{\mathrm{\α}}_i^{{*}^T}{K}_M{\mathrm{\α}}_{\mathrm{if}}{\mathrm{\α}}_{\mathrm{if}}^T\mathrm{\Φ}{\left({\hat{X}}_f\right)}^T\end{array}---\left(27\right)$

$\begin{array}{c}{E}_{\mathrm{nf}}^{{*}^T}=E^T{P}_{x,f}^{*}{P}_{x,f}^{{*}^T}=\mathrm{\Φ}{\left(\hat{X}\right)}^T{P}_x^{*}{P}_x^{{*}^T}{P}_{x,f}^{*}{P}_{x,f}^{{*}^T}\\ =K_i{\mathrm{\α}}_i^{*}{\mathrm{\α}}_i^{{*}^T}\mathrm{\Φ}{\left(\hat{X}\right)}^T\mathrm{\Φ}\left({\hat{X}}_f\right){\mathrm{\α}}_{\mathrm{if}}^{*}{\mathrm{\α}}_{\mathrm{if}}^{{*}^T}\mathrm{\Φ}{\left({\hat{X}}_f\right)}^T\\ =K_i{\mathrm{\α}}_i^{*}{\mathrm{\α}}_i^{{*}^T}{K}_M{\mathrm{\α}}_{\mathrm{if}}^{*}{\mathrm{\α}}_{\mathrm{if}}^{{*}^T}\mathrm{\Φ}{\left({\hat{X}}_f\right)}^T\end{array}---\left(28\right)$

Step 5.6.2: utilize kernel principal component analysis to determine the data set of the principal component space in the residual error space in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line after orthogonal mapping pivot T _edata set with the residual error space in the residual error space in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line after orthogonal mapping pivot

$\begin{array}{c}{T}_e=E_{\mathrm{nf}}^T{P}_e=K_i{\mathrm{\α}}_i^{{*}^T}{K}_M{\mathrm{\α}}_{\mathrm{if}}{\mathrm{\α}}_{\mathrm{if}}^T\mathrm{\Φ}{\left({\hat{X}}_f\right)}^T\mathrm{\Φ}\left({\hat{X}}_f\right)\\ {\mathrm{\α}}_{\mathrm{if}}{\mathrm{\α}}_{\mathrm{if}}^T{K}_M^T{\mathrm{\α}}_i^{*}{\mathrm{\α}}_i^{{*}^T}{K}_i^T{\mathrm{\α}}_{\mathrm{nf}}\\ =K_i{\mathrm{\α}}_i^{*}{\mathrm{\α}}_i^{{*}^T}{K}_M{\mathrm{\α}}_{\mathrm{if}}{\mathrm{\α}}_{\mathrm{if}}^T{K}_{\mathrm{if}}{\mathrm{\α}}_{\mathrm{if}}{\mathrm{\α}}_{\mathrm{if}}^T{K}_M^T{\mathrm{\α}}_i^{*}{\mathrm{\α}}_i^{{*}^T}{K}_i^T{\mathrm{\α}}_{\mathrm{nf}}\end{array}---\left(29\right)$

$\begin{array}{c}{T}_e^{*}=E_{\mathrm{nf}}^{{*}^T}{P_e}^{*}\\ =K_i{\mathrm{\α}}_i^{*}{\mathrm{\α}}_i^{{*}^T}{K}_M{\mathrm{\α}}_{\mathrm{if}}^{*}{\mathrm{\α}}_{\mathrm{if}}^{{*}^T}{K}_{\mathrm{if}}{\mathrm{\α}}_{\mathrm{if}}^{*}{\mathrm{\α}}_{\mathrm{if}}^{{*}^T}{K}_M^T{\mathrm{\α}}_i^{*}{\mathrm{\α}}_i^{{*}^T}{K}_i^T{\mathrm{\α}}_{\mathrm{nf}}^{*}\end{array}---\left(30\right)$

In formula, α _nffor Ε _nfthe proper vector of principal component space, for Ε _nf ^*the proper vector of principal component space, P _efor E _nfthe load matrix of principal component space, for Ε _nf ^*the load matrix of principal component space, P _e=E _nfα _nf,

Step 5.6.3: the data set of the principal component space in the residual error space of penicillin fermentation process performance variable off-line historical failure data set be mapped to the load matrix P of the principal component space in the residual error space in the residual error space of the penicillin fermentation process performance variable off-line historical failure data set after orthogonal mapping _e, the data set E in the residual error space in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line _f ^tbe mapped to the load matrix in the residual error space in the residual error space in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line after orthogonal mapping obtain the data set of the mapping space of the principal component space in the residual error space in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line data set E with the mapping space in the residual error space in the residual error space in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line _f,e;

$\begin{array}{c}{E}_{f,e}={E_f}^T{P}_e^{*}{P}_e^{{*}^T}=K_{\mathrm{if}}{\mathrm{\α}}_{\mathrm{if}}^{*}{\mathrm{\α}}_{\mathrm{if}}^{{*}^T}{K}_M{\mathrm{\α}}_{\mathrm{if}}^{*}{\mathrm{\α}}_{\mathrm{if}}^{{*}^T}{K}_M^T\\ {\mathrm{\α}}_i^{*}{\mathrm{\α}}_i^{{*}^T}{K}_i^T{\mathrm{\α}}_{\mathrm{nf}}^{*}{\mathrm{\α}}_{\mathrm{nf}}^{{*}^T}{K}_i{\mathrm{\α}}_i^{*}{\mathrm{\α}}_i^{{*}^T}{K}_M{\mathrm{\α}}_{\mathrm{if}}^{*}{\mathrm{\α}}_{\mathrm{if}}^{{*}^T}\mathrm{\Φ}{\left({\hat{X}}_f\right)}^T\end{array}---\left(32\right)$

Step 5.6.4: calculate penicillin fermentation process performance variable off-line historical failure data set Χ _fthe data set of principal component space in residual error space arrive the data set of the mapping space of the principal component space in the residual error space in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line the distance of load matrix, calculate the data set E in residual error space in the residual error space of the historical normal data set Χ of penicillin fermentation process performance variable off-line to the data set of the mapping space in the residual error space in the residual error space in the residual error space of penicillin fermentation process performance variable off-line history normal data set the distance of load matrix, and calculate the ratio R T of two distances _f, _e;

Step 5.6.5: calculate penicillin fermentation process performance variable off-line historical failure data set Χ _fthe data set E in residual error space in residual error space _farrive the data set of the mapping space of the principal component space in the residual error space in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line the distance of load matrix, calculate the data set E in residual error space in the residual error space of the historical normal data set Χ of penicillin fermentation process performance variable off-line to the data set of the mapping space in the residual error space in the residual error space in the residual error space of penicillin fermentation process performance variable off-line history normal data set the distance of load matrix, and calculate the ratio of two distances

${\mathrm{RT}}_{f,e}^{*}=\frac{{\left|\right|E_f{P}_{f,e}^{*}{P}_{f,e}^{{*}^T}\left|\right|}^2}{{\left|\right|{\mathrm{EP}}_{f,e}^{*}{P}_{f,e}^{{*}^T}\left|\right|}^2},(e=\mathrm{1,2},\·\·\·,\mathrm{Be}*)---\left(34\right)$

In formula, || || represent Euclid's length, P _f,efor mapping space load matrix, for mapping space E _f,eload matrix, this ratio of distances constant is larger, represents the residual error space of penicillin fermentation process performance variable off-line historical failure data set larger to the contribution of SPE statistic along each load direction;

Step 5.6.6: the ratio of this two classes distance is greater than the pivot direction V that the principal component space in residual error space of penicillin fermentation process performance variable off-line historical failure data set that the ratio of its all distances of setting ratio lower limit is corresponding and the residual error space in the residual error space of penicillin fermentation process performance variable off-line historical failure data set form respectively the principal component space in the residual error space in the residual error space of new penicillin fermentation process performance variable off-line historical failure data set _fepivot direction with the residual error space in the residual error space in the residual error space of new penicillin fermentation process performance variable off-line historical failure data set

Step 5.6.7: the pivot direction V that utilizes the principal component space in the residual error space in the residual error space of new penicillin fermentation process performance variable off-line historical failure data set _fepivot direction with the residual error space in the residual error space in the residual error space of new penicillin fermentation process performance variable off-line historical failure data set the fault related direction P in the residual error space in the residual error space of the historical normal data set of reconstruct penicillin fermentation process performance variable off-line _{rec, e};

$P_{\mathrm{rec},e}=[V_{\mathrm{fe}},V_{\mathrm{fe}}^{*}]=[E_{\mathrm{nf}}{\mathrm{\α}}_{\mathrm{fe}},E_{\mathrm{nf}}^{*},{\mathrm{\α}}_{\mathrm{fe}}^{*}]---\left(35\right)$

Wherein, α _fefor pivot direction V _fecharacteristic of correspondence vector, for pivot direction characteristic of correspondence vector;

Step 5.7: for all faults of penicillin fermentation process, applying step 5.1-5.6 reconstructs respectively the fault related direction P of the principal component space of penicillin fermentation process performance variable off-line historical failure data set _{rec, c}, penicillin fermentation process performance variable off-line historical failure data set the fault related direction P of principal component space in residual error space _{rec, p}, the historical normal data set of penicillin fermentation process performance variable off-line the fault related direction P in residual error space in residual error space _{rec, e}form the penicillin fermentation process fault related direction based on the reconstruct of improved core offset minimum binary, thereby set up the model bank of the penicillin fermentation process fault related direction of improved core offset minimum binary reconstruct;

Step 6: utilize the relevant direction model of penicillin fermentation process fault to carry out penicillin fermentation process fault diagnosis;

Step 6.1: according to the fault related direction P of the principal component space of penicillin fermentation process performance variable off-line historical failure data set _{rec, c}, the score matrix Τ of the principal component space of the penicillin fermentation process performance variable data that calculating is obtained online _foand corresponding Hotelling statistic

$\left\{\begin{array}{c}\mathrm{\Φ}{\left({\tilde{X}}_{f,\mathrm{rec},\mathrm{new}}\right)}^T=\mathrm{\Φ}{\left({\tilde{X}}_{f,\mathrm{new}}\right)}^T{P}_{\mathrm{rec},c}{P}_{\mathrm{rec},c}^T\\ \mathrm{\Φ}{\left({\tilde{X}}_{\mathrm{fo}}\right)}^T=\mathrm{\Φ}{\left({\tilde{X}}_{f,\mathrm{new}}\right)}^T-\mathrm{\Φ}{\left({\tilde{X}}_{f,\mathrm{rec},\mathrm{new}}\right)}^T\end{array}\right.---\left(36\right)$

$\begin{array}{c}{T}_{\mathrm{fo}}=\mathrm{\Φ}{\left({\tilde{X}}_{\mathrm{fo}}\right)}^{T}P\\ =\mathrm{\Φ}{\left({\tilde{X}}_{f,\mathrm{new}}\right)}^{T}P-\mathrm{\Φ}{\left({\tilde{X}}_{f,\mathrm{rec},\mathrm{new}}\right)}^{T}P\\ =K_{f,\mathrm{new}}{\mathrm{\α}}_{f,\mathrm{new}}{\mathrm{\α}}_{f,\mathrm{new}}^T\mathrm{\Φ}{\left(X_{f,\mathrm{new}}\right)}^T\mathrm{\Φ}\left(X\right)\mathrm{\α}-\mathrm{\Φ}{\left({\tilde{X}}_{f,\mathrm{new}}\right)}^T{P}_{\mathrm{rec},c}{P}_{\mathrm{rec},c}^T\mathrm{\Φ}\left(X\right)\mathrm{\α}\\ =K_{f,\mathrm{new}}{\mathrm{\α}}_{f,\mathrm{new}}{\mathrm{\α}}_{f,\mathrm{new}}^T{K}_{N,\mathrm{new}}^T\mathrm{\α}-\mathrm{\Φ}{\left({\tilde{X}}_{f,\mathrm{new}}\right)}^T\mathrm{\Φ}\left(X_f\right){\mathrm{\α}}_f{\mathrm{\α}}_f^T{K}_N^T{\mathrm{\α\α}}^T{K}^T{\mathrm{\α}}_{\mathrm{fc}}{\mathrm{\α}}_{\mathrm{fc}}^{T}K{\mathrm{\α\α}}^T{K}_N{\mathrm{\α}}_f{{\mathrm{\α}}_f}^T\mathrm{\Φ}{\left(X_f\right)}^T\mathrm{\Φ}\left(X\right)\mathrm{\α}\\ -\mathrm{\Φ}{\left(\stackrel{}{{\tilde{X}}_{f,\mathrm{new}}}\right)}^T{\mathrm{\Φ}}_f\left(X\right){\mathrm{\α}}_f^{*}{\mathrm{\α}}_f^{{*}^T}{K}_N^T{\mathrm{\α\α}}^T{K}^T{\mathrm{\α}}_{\mathrm{fc}}^{*}{\mathrm{\α}}_{\mathrm{fc}}^{{*}^T}K{\mathrm{\α\α}}^T{K}_N{\mathrm{\α}}_f^{*}{\mathrm{\α}}_f^{{*}^T}\mathrm{\Φ}{\left(X_f\right)}^T\mathrm{\Φ}\left(X\right)\mathrm{\α}\\ =K_{f,\mathrm{new}}{\mathrm{\α}}_{f,\mathrm{new}}{\mathrm{\α}}_{f,\mathrm{new}}^T{K}_{N,\mathrm{new}}^T\mathrm{\α}-K_{f,c,\mathrm{new}}{\mathrm{\α}}_f{\mathrm{\α}}_f^T{K}_N^T{\mathrm{\α\α}}^T{K}^T{\mathrm{\α}}_{\mathrm{fc}}{\mathrm{\α}}_{\mathrm{fc}}^{T}K{\mathrm{\α\α}}^T{K}_N{\mathrm{\α}}_f{{\mathrm{\α}}_f}^T{K}_f\mathrm{\α}\\ -K_{f,c,\mathrm{new}}{\mathrm{\α}}_f^{*}{\mathrm{\α}}_f^{{*}^T}{K}_N^T{\mathrm{\α\α}}^T{K}^T{\mathrm{\α}}_{\mathrm{fc}}^{*}{\mathrm{\α}}_{\mathrm{fc}}^{{*}^T}K{\mathrm{\α\α}}^T{K}_N{\mathrm{\α}}_f^{*}{\mathrm{\α}}_f^{{*}^T}{K}_f\mathrm{\α}\end{array}---\left(37\right)$

$T_{\mathrm{fo},i}^2=t_{\mathrm{fo},i}^T{\mathrm{\Λ}}_{\mathrm{fo}}^{-1}{t}_{\mathrm{fo},i},(i=\mathrm{1,2},\·\·\·,J_{\mathrm{new}})---\left(38\right)$

In formula, for x _newthe set of principal component space, t _fofor score matrix Τ _forow vector, Λ _fofor the covariance matrix of the principal component space of penicillin fermentation process performance variable off-line historical failure data set, J _newfor the number of samples of the penicillin fermentation process performance variable data obtained online, K _{f, new}for the penicillin fermentation process performance variable data x obtaining online _newset X _newwith the penicillin fermentation process performance variable data x obtaining online _newset X _newinterior product matrix, K _{n, new}for the historical normal data set Χ of penicillin fermentation process performance variable off-line and the penicillin fermentation process performance variable data x obtaining online _newset X _newinterior product matrix, K _{f, c, new}for the penicillin fermentation process performance variable data x obtaining online _newset X _newprincipal component space data set and penicillin fermentation process performance variable off-line historical failure data set Χ _finterior product matrix, α _{f, new}for Φ (X _{f, new}) the proper vector of principal component space;

Step 6.2: according to the fault related direction of the principal component space in the residual error space of penicillin fermentation process performance variable off-line historical failure data set, calculate the score matrix Τ in the residual error space of the penicillin fermentation process performance variable data of obtaining online _{x, fo}and corresponding Hotelling statistic

$T_{x,\mathrm{fo},i}^2=t_{x,\mathrm{fo},i}^T{\mathrm{\Λ}}_{x,\mathrm{fo}}^{-1}{t}_{x,\mathrm{fo},i},(i=\mathrm{1,2},\·\·\·,J_{\mathrm{new}})---\left(41\right)$

In formula, t _{x, fo}for score matrix Τ _{x, fo}row vector, Λ _{x, fo}for the covariance matrix in the residual error space of penicillin fermentation process performance variable off-line historical failure data set, K _{if, new}for the penicillin fermentation process performance variable data x obtaining online _newset X _newresidual error space data sets with the penicillin fermentation process performance variable data x obtaining online _newset X _newresidual error space data sets interior product matrix, K _{i, new}for the residual error space data sets of the historical normal data set Χ of penicillin operational process off-line with the penicillin fermentation process performance variable data x obtaining online _newset X _newresidual error space data sets interior product matrix, K _{f, p, new}for the penicillin fermentation process performance variable data x obtaining online _newset X _newthe principal component space data set in residual error space with penicillin fermentation process performance variable off-line historical failure data set Χ _fthe data in residual error space interior product matrix, α _{if, new}for the proper vector of principal component space;

Step 6.3: according to the barrier related direction in the residual error space in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line, calculate the SPE statistic in the residual error space of the penicillin fermentation process performance variable data of obtaining online;

${\mathrm{SPE}}_{x,\mathrm{fo},i}=e_{x,\mathrm{fo}.i}^T{e}_{x,\mathrm{fo},i},(i=\mathrm{1,2},\·\·\·,J_{\mathrm{new}})---\left(44\right)$

In formula, e _{x, fo}for score matrix Ε _{x, fo}row vector;

Step 6.4: the Hotelling statistic of the principal component space of the penicillin fermentation process performance variable data that judgement is obtained online the Hotelling statistic in the residual error space of the penicillin fermentation process performance variable data of obtaining online whether the SPE statistic in the residual error space of the penicillin fermentation process performance variable data of obtaining is online all lower than the confidence limit separately of setting, be,, when prior fault is fault real in penicillin fermentation process, be out of order combination or new fault otherwise work as prior fault.

In present embodiment, first choose 300 sampled points of the historical normal data of off-line of penicillin fermentation process, utilize improved core offset minimum binary method to set up the malfunction monitoring model of penicillin fermentation process, then on-line monitoring fault type 1 and fault type 2, the data of table 1～table 5 are partial data:

Table 1 is set up four groups of data in penicillin fermentation process malfunction monitoring model

Table 2 is set up four groups of data in the fault data 1 in the relevant direction model of penicillin fermentation process fault of improved core offset minimum binary reconstruct

Table 3 is set up four groups of data in the fault data 2 in the relevant direction model of penicillin fermentation process fault of improved core offset minimum binary reconstruct

Table 4 is monitored one group of data in the fault data 1 of penicillin fermentation process

Table 5 is monitored one group of data in the fault data 2 of penicillin fermentation process

Can find out the Hotelling statistic of the performance variable principal component space in penicillin fermentation process and the Hotelling statistic in process variable residual error space by Fig. 2 and Fig. 3 in 200 points, all do not exceed confidence limit with SPE statistic, since the 300th sampled point, there is significantly transfiniting phenomenon in above-mentioned three statistics, and this is true to life.Illustrate that the penicillin fermentation process method for diagnosing faults based on the reconstruct of core offset minimum binary can be good at diagnosing out the fault in Penicillium sweat.

Then, the reconstruct fault related direction that 300 history samples points of fault type 1 carry out, inline diagnosis fault type 1, as seen from Figure 4, the Hotelling statistic in the Hotelling statistic of the principal component space of the performance variable in penicillin fermentation process and the residual error space of performance variable all lower than confidence limit, illustrate that fault type 1 is for current fault with SPE statistic, and the relevant direction model of this fault can be good at diagnosis and is out of order.

With the fault related direction on-line monitoring fault type 2 of 300 sampled point reconstruct of fault type 1, as seen from Figure 7, the Hotelling statistic (Hotelling-T of the principal component space of the performance variable in penicillin fermentation process ²) and the Hotelling statistic in the residual error space of performance variable all higher than confidence limit, the relevant direction model of the fault of application fault type 1 can not be diagnosed fault type 2 with SPE statistic.

Utilize equally 300 sampled points of fault type 2 to be reconstructed fault related direction, inline diagnosis fault type 2, as seen from Figure 6, the Hotelling statistic in the Hotelling statistic of the principal component space of the performance variable in penicillin fermentation process and the residual error space of performance variable all lower than confidence limit, explanation fault type 1 is current fault with SPE statistic, and the relevant direction model of this fault can be good at diagnosis and is out of order.

Next, use the fault related direction on-line monitoring fault type 1 of 300 sampled point reconstruct of fault type 2, as seen from Figure 5, the Hotelling statistic (Hotelling-T of the principal component space of the performance variable in penicillin fermentation process ²) and the Hotelling statistic in the residual error space of performance variable all higher than confidence limit, the relevant direction model of the fault of application fault type 2 can not be diagnosed fault type 1 with SPE statistic.

Comparison diagram 4 and Fig. 5 or (Fig. 6 and Fig. 7) can find out that the present invention is to diagnosing the different faults of same performance variable.

Can be obtained by the above results, by penicillin fermentation process method for diagnosing faults of the present invention, can effectively monitor the different faults type of same process variable.

Although more than described the specific embodiment of the present invention, the those skilled in the art in this area should be appreciated that these only illustrate, and can make various changes or modifications to these embodiments, and not deviate from principle of the present invention and essence.Scope of the present invention is only limited by appended claims.

Claims (4)

1. the penicillin fermentation process method for diagnosing faults based on the reconstruct of core offset minimum binary, is characterized in that: comprise the following steps:

Step 1: gather the historical normal data of off-line of penicillin fermentation process, comprise the historical normal data set of penicillin fermentation process performance variable off-line and the historical normal data set of penicillin fermentation process state variable off-line;

Penicillin fermentation process performance variable comprises the temperature of temperature, pH value and the fermentation reactor of flow velocity, the substrate feeding of ventilation rate, agitator power, substrate feeding;

Penicillin fermentation process state variable comprises the concentration of penicillin, the heat that penicillin reaction produces, concentration, the volume of nutrient culture media and the concentration of thalline of carbon dioxide;

Step 2: respectively the historical normal data set of penicillin fermentation process performance variable off-line and the historical normal data set of penicillin fermentation process state variable off-line are carried out to specification and standardization;

Step 3: utilize improved core offset minimum binary method to set up the malfunction monitoring model of penicillin fermentation process, this model be input as the historical normal data set of penicillin fermentation process performance variable off-line, this model is output as the historical normal data set of penicillin fermentation process state variable off-line;

Step 3.1: data core mapping: historical penicillin fermentation process performance variable off-line normal data set is mapped to high-dimensional feature space by kernel function from original data space;

Step 3.2: utilize kernel partial least squares that historical penicillin fermentation process performance variable off-line normal data set and the historical normal data set of penicillin fermentation process state variable off-line are divided into respectively to principal component space and residual error space;

Step 3.3: utilize principal component space that kernel principal component analysis is residual error space by the residual error spatial division of historical penicillin fermentation process performance variable off-line normal data set and the residual error space in residual error space, and then obtain the malfunction monitoring model of penicillin fermentation process:

Historical penicillin fermentation process performance variable off-line normal data set is expressed as to data set, the data set of principal component space in residual error space and the data set sum in the residual error space in residual error space of its principal component space, historical penicillin fermentation process state variable off-line normal data set is expressed as to the data set of its principal component space and the data set sum in residual error space;

Step 4: utilize the malfunction monitoring model of penicillin fermentation process, the fault of on-line monitoring penicillin fermentation process;

Step 4.1: obtain online penicillin fermentation process performance variable data and penicillin fermentation process state variable data;

Step 4.2: the penicillin fermentation process performance variable data of obtaining online and penicillin fermentation process state variable data are carried out to specification and standardization;

Step 4.3: data core mapping: the penicillin fermentation process performance variable data of obtaining are online mapped to high-dimensional feature space by kernel function from original data space;

Step 4.4: calculate Hotelling statistic in the principal component space of the historical normal data set of penicillin fermentation process performance variable off-line of the penicillin fermentation process performance variable data obtained online, online Hotelling statistic in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line of the penicillin fermentation process performance variable data obtained and the penicillin fermentation process performance variable data obtained the online SPE statistic in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line;

Step 4.5: the Hotelling statistic of the penicillin fermentation process performance variable data of obtaining online that judgement calculates in the principal component space of the historical normal data set of penicillin fermentation process performance variable off-line, in the Hotelling statistic of the penicillin fermentation process performance variable data of obtaining online in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line and the penicillin fermentation process performance variable data obtained the online SPE statistic in the residual error space of penicillin fermentation process performance variable off-line history normal data set, whether there is at least one statistic higher than its corresponding confidence limit: be, current penicillin fermentation process breaks down, execution step 5, otherwise return to step 4.1,

Step 5: set up the relevant direction model of penicillin fermentation process fault based on the reconstruct of improved core offset minimum binary;

Step 5.1: gather the off-line historical failure data of penicillin fermentation process, comprise penicillin fermentation process performance variable off-line historical failure data set and penicillin fermentation process state variable off-line historical failure data set;

Step 5.2: penicillin fermentation process performance variable off-line historical failure data set and penicillin fermentation process state variable off-line historical failure data set are carried out to specification and standardization;

Step 5.3: penicillin fermentation process performance variable off-line historical failure data set is mapped to high-dimensional feature space by kernel function from original data space;

Step 5.4: according to the historical normal data set of penicillin fermentation process performance variable off-line and penicillin fermentation process performance variable off-line historical failure data set, the fault related direction of the principal component space of reconstruct penicillin fermentation process performance variable off-line historical failure data set;

Step 5.5: according to the fault related direction of the principal component space in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line and penicillin fermentation process performance variable off-line historical failure Data set reconstruction penicillin fermentation process performance variable off-line historical failure data set;

Step 5.6: according to the fault related direction in the residual error space in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line and penicillin fermentation process performance variable off-line historical failure Data set reconstruction penicillin fermentation process performance variable off-line historical failure data set;

Step 5.7: for all faults of penicillin fermentation process, applying step 5.1-5.6 reconstructs respectively the fault related direction of the principal component space of penicillin fermentation process performance variable off-line historical failure data set, the fault related direction of the principal component space in the residual error space of penicillin fermentation process performance variable off-line historical failure data set, the fault related direction in the residual error space in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line forms the penicillin fermentation process fault related direction based on the reconstruct of improved core offset minimum binary, thereby set up the model bank of the penicillin fermentation process fault related direction of improved core offset minimum binary reconstruct,

Step 6: utilize the relevant direction model of penicillin fermentation process fault to carry out penicillin fermentation process fault diagnosis;

Step 6.1: according to the fault related direction of the principal component space of penicillin fermentation process performance variable off-line historical failure data set, calculate score matrix and the corresponding Hotelling statistic thereof of the principal component space of the penicillin fermentation process performance variable data of obtaining online;

Step 6.2: according to the fault related direction of the principal component space in the residual error space of penicillin fermentation process performance variable off-line historical failure data set, calculate score matrix and the corresponding Hotelling statistic thereof in the residual error space of the penicillin fermentation process performance variable data of obtaining online;

Step 6.3: according to the fault related direction in the residual error space in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line, calculate the SPE statistic in the residual error space of the penicillin fermentation process performance variable data of obtaining online;

Step 6.4: the Hotelling statistic of the principal component space of the penicillin fermentation process performance variable data obtained online of judgement, online the residual error space of the penicillin fermentation process performance variable data obtained Hotelling statistic, whether the SPE statistic in the residual error space of the penicillin fermentation process performance variable data obtained all lower than the confidence limit separately of setting online, be,, when prior fault is fault real in penicillin fermentation process, be out of order combination or new fault otherwise work as prior fault.

2. the penicillin fermentation process method for diagnosing faults based on the reconstruct of core offset minimum binary according to claim 1, is characterized in that: described step 5.4 is specifically carried out according to the following steps:

Step 5.4.1: utilize core offset minimum binary method that penicillin fermentation process performance variable off-line historical failure data set and the historical normal data set of penicillin fermentation process performance variable off-line are divided respectively to principal component space and residual error space, obtain respectively the core partial least square model of penicillin fermentation process performance variable off-line historical failure data set and the core partial least square model of the historical normal data set of penicillin fermentation process performance variable off-line;

Step 5.4.2: the data set of the principal component space of historical penicillin fermentation process performance variable off-line normal data set respectively orthogonal mapping to the load matrix in the load matrix of the principal component space of penicillin operational process performance variable off-line historical failure data set and the residual error space of penicillin operational process performance variable off-line historical failure data set, obtain the data set in the residual error space of the principal component space of the penicillin fermentation process performance variable off-line history normal data set after data set and the orthogonal mapping of principal component space of the principal component space of the historical normal data set of penicillin fermentation process performance variable off-line after orthogonal mapping;

Step 5.4.3: utilize kernel principal component analysis to determine the pivot of the data set in the residual error space of the principal component space of the historical normal data set of penicillin fermentation process performance variable off-line after pivot and the orthogonal mapping of data set of the principal component space of the principal component space of the historical normal data set of penicillin fermentation process performance variable off-line after orthogonal mapping;

Step 5.4.4: the pivot in the load matrix direction in the residual error space of the principal component space of the historical normal data set of penicillin fermentation process performance variable off-line of the data set in the pivot in the load matrix direction of the principal component space of the principal component space of the historical normal data set of penicillin fermentation process performance variable off-line of the data set of principal component space of asking respectively penicillin fermentation process performance variable off-line historical failure data set after orthogonal mapping and the residual error space of penicillin fermentation process performance variable off-line historical failure data set after orthogonal mapping, obtains respectively the fault pivot of principal component space of penicillin fermentation process performance variable off-line historical failure data set and the fault pivot in the residual error space of penicillin fermentation process performance variable off-line historical failure data set;

Step 5.4.5: proportion in the pivot of the principal component space of the principal component space of the historical normal data set of penicillin fermentation process performance variable off-line of the fault pivot of the principal component space of calculating penicillin fermentation process performance variable off-line historical failure data set after orthogonal mapping, proportion in the pivot in the residual error space of the principal component space of the historical normal data set of penicillin fermentation process performance variable off-line of the fault pivot in the residual error space of calculating penicillin fermentation process performance variable off-line historical failure data set after orthogonal mapping;

Step 5.4.6: the fault pivot that is greater than the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line after fault pivot and the orthogonal mapping of principal component space of the corresponding penicillin fermentation process performance variable of the each ratio value off-line historical failure data set of the ratio lower limit of setting in two class ratio values forms respectively the pivot direction of principal component space of new penicillin fermentation process performance variable off-line historical failure data set and the pivot direction in the residual error space of new penicillin fermentation process performance variable off-line historical failure data set;

Step 5.4.7: load matrix corresponding to pivot direction of utilizing the residual error space of load matrix corresponding to the pivot direction of principal component space of new penicillin fermentation process performance variable off-line historical failure data set and new penicillin fermentation process performance variable off-line historical failure data set, the load matrix in the residual error space of the principal component space of the load matrix of the principal component space of the principal component space of the historical normal data set of reconstruct penicillin fermentation process performance variable off-line and the historical normal data set of penicillin fermentation process performance variable off-line, i.e. the fault related direction of the principal component space of penicillin fermentation process performance variable off-line historical failure data set.

3. the penicillin fermentation process method for diagnosing faults based on the reconstruct of core offset minimum binary according to claim 1, is characterized in that: described step 5.5 is specifically carried out according to the following steps:

Step 5.5.1: utilize core pivot element analysis method, be the principal component space in residual error space and the residual error space in residual error space by the residual error spatial division of the residual error space of penicillin fermentation process performance variable off-line historical failure data set and the historical normal data set of penicillin fermentation process performance variable off-line respectively, obtain respectively the kernel pivot model in residual error space of penicillin fermentation process performance variable off-line historical failure data set and the kernel pivot model in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line;

Step 5.5.2: the data set of the principal component space in the residual error space of historical penicillin fermentation process performance variable off-line normal data set respectively orthogonal mapping to the load matrix in the residual error space in the load matrix of the principal component space in the residual error space of penicillin fermentation process performance variable off-line historical failure data set and the residual error space of penicillin fermentation process performance variable off-line historical failure data set, obtain the data set in the residual error space in the residual error space of the penicillin fermentation process performance variable off-line history normal data set after data set and the orthogonal mapping of principal component space in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line after orthogonal mapping;

Step 5.5.3: utilize kernel principal component analysis to determine the pivot of the data set in the residual error space in the residual error space of the penicillin fermentation process performance variable off-line historical failure data set after pivot and the orthogonal mapping of data set of the principal component space in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line after orthogonal mapping;

Step 5.5.4: the pivot in the load matrix direction of the principal component space in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line of the data set of principal component space of asking the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line after orthogonal mapping, pivot in the load matrix direction in the residual error space in the residual error space of the penicillin fermentation process performance variable off-line historical failure data set of the data set in residual error space of asking the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line after orthogonal mapping, it is the fault pivot in the fault pivot of principal component space in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line and the residual error space in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line,

Step 5.5.5: proportion in the pivot of the principal component space in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line of the fault pivot of the principal component space in the residual error space of the historical normal data set of calculating penicillin fermentation process performance variable off-line after orthogonal mapping, proportion in the pivot of fault pivot principal component space in the residual error space of penicillin fermentation process performance variable off-line historical failure data set after orthogonal mapping in the residual error space in the residual error space of the historical normal data set of calculating penicillin fermentation process performance variable off-line;

Step 5.5.6: the fault pivot that is greater than the residual error space in the fault pivot of principal component space in the residual error space of the corresponding penicillin fermentation process performance variable of the each ratio value off-line historical failure data set of the ratio lower limit of setting and the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line in two class ratio values forms respectively the pivot direction in the pivot direction of principal component space in the residual error space of new penicillin fermentation process performance variable off-line historical failure data set and the residual error space in the residual error space of new penicillin fermentation process performance variable off-line historical failure data set;

Step 5.5.7: load matrix corresponding to pivot direction of utilizing the residual error space in the residual error space of load matrix corresponding to the pivot direction of principal component space in the residual error space of new penicillin fermentation process performance variable off-line historical failure data set and new penicillin fermentation process performance variable off-line historical failure data set, the load matrix in the residual error space in the load matrix of the principal component space in the residual error space of reconstruct penicillin fermentation process performance variable off-line historical failure data set and the residual error space of penicillin fermentation process performance variable off-line historical failure data set, i.e. the fault related direction of the principal component space in the residual error space of penicillin fermentation process performance variable off-line historical failure data set.

4. the penicillin fermentation process method for diagnosing faults based on the reconstruct of core offset minimum binary according to claim 1, is characterized in that: described step 5.6 is specifically carried out according to the following steps:

Step 5.6.1: by the data set in the residual error space in the residual error space of historical penicillin fermentation process performance variable off-line normal data set respectively orthogonal mapping to the load matrix in the residual error space in the load matrix of the principal component space in the residual error space of penicillin fermentation process performance variable off-line historical failure data set and the residual error space of penicillin fermentation process performance variable off-line historical failure data set, obtain the residual error space in the residual error space in the residual error space of the penicillin fermentation process performance variable off-line history normal data set after principal component space and the orthogonal mapping in residual error space in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line after orthogonal mapping;

Step 5.6.2: utilize kernel principal component analysis to determine the pivot of the data set in the residual error space in the residual error space in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line after pivot and the orthogonal mapping of data set of the principal component space in the residual error space in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line after orthogonal mapping;

Step 5.6.3: the load matrix that the data set of the principal component space in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line is mapped to the principal component space in the residual error space in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line after orthogonal mapping, the data set in the residual error space in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line is mapped to the load matrix in the residual error space in the residual error space in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line after orthogonal mapping, obtain the data set of the mapping space in the residual error space in the data set of mapping space of the principal component space in the residual error space in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line and the residual error space in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line,

Step 5.6.4: the data set of the principal component space in the residual error space of calculating penicillin fermentation process performance variable off-line historical failure data set is to the distance of the load matrix of the data set of the mapping space of the principal component space in the residual error space in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line, the data set in the residual error space in the residual error space of the historical normal data set of calculating penicillin fermentation process performance variable off-line is to the distance of the load matrix of the data set of the mapping space in the residual error space in the residual error space in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line, and calculate the ratio of two distances,

Step 5.6.5: the data set in the residual error space in the residual error space of calculating penicillin fermentation process performance variable off-line historical failure data set is to the distance of the load matrix of the data set of the mapping space of the principal component space in the residual error space in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line, the data set in the residual error space in the residual error space of the historical normal data set of calculating penicillin fermentation process performance variable off-line is to the distance of the load matrix of the data set of the mapping space in the residual error space in the residual error space in the residual error space of the historical normal data set of penicillin fermentation process performance variable off-line, and calculate the ratio of two distances,

Step 5.6.6: the ratio of this two classes distance is greater than the pivot direction that the principal component space in residual error space of penicillin fermentation process performance variable off-line historical failure data set that the ratio of its all distances of setting ratio lower limit is corresponding and the residual error space in the residual error space of penicillin fermentation process performance variable off-line historical failure data set form respectively the residual error space in the pivot direction of principal component space in the residual error space in the residual error space of new penicillin fermentation process performance variable off-line historical failure data set and the residual error space in the residual error space of new penicillin fermentation process performance variable off-line historical failure data set;

Step 5.6.7: utilize the pivot direction in the residual error space in the pivot direction of principal component space in the residual error space in the residual error space of new penicillin fermentation process performance variable off-line historical failure data set and the residual error space in the residual error space of new penicillin fermentation process performance variable off-line historical failure data set, the fault related direction in the residual error space in the residual error space of the historical normal data set of reconstruct penicillin fermentation process performance variable off-line.