CN111598435B - Quality trend prediction method based on self-adaptive feature selection and improved thinking evolutionary algorithm - Google Patents
Quality trend prediction method based on self-adaptive feature selection and improved thinking evolutionary algorithm Download PDFInfo
- Publication number
- CN111598435B CN111598435B CN202010405648.2A CN202010405648A CN111598435B CN 111598435 B CN111598435 B CN 111598435B CN 202010405648 A CN202010405648 A CN 202010405648A CN 111598435 B CN111598435 B CN 111598435B
- Authority
- CN
- China
- Prior art keywords
- data
- particles
- mode
- quality
- subgroup
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 55
- 238000004422 calculation algorithm Methods 0.000 title claims abstract description 30
- 230000004927 fusion Effects 0.000 claims abstract description 14
- 238000012545 processing Methods 0.000 claims abstract description 7
- 239000002245 particle Substances 0.000 claims description 110
- 230000009467 reduction Effects 0.000 claims description 9
- 230000008859 change Effects 0.000 claims description 8
- 230000000737 periodic effect Effects 0.000 claims description 8
- 208000036762 Acute promyelocytic leukaemia Diseases 0.000 claims description 6
- 238000013528 artificial neural network Methods 0.000 claims description 6
- 208000003343 Antiphospholipid Syndrome Diseases 0.000 claims description 5
- 238000007781 pre-processing Methods 0.000 claims description 5
- 230000009191 jumping Effects 0.000 claims description 3
- 238000005457 optimization Methods 0.000 claims description 3
- 230000001172 regenerating effect Effects 0.000 claims description 3
- 238000004088 simulation Methods 0.000 claims description 2
- 239000008186 active pharmaceutical agent Substances 0.000 claims 1
- 230000001174 ascending effect Effects 0.000 claims 1
- 238000007500 overflow downdraw method Methods 0.000 abstract description 3
- 238000004519 manufacturing process Methods 0.000 description 25
- 239000011159 matrix material Substances 0.000 description 14
- 230000008569 process Effects 0.000 description 11
- 230000002159 abnormal effect Effects 0.000 description 9
- 238000004364 calculation method Methods 0.000 description 6
- 238000009826 distribution Methods 0.000 description 5
- 230000006870 function Effects 0.000 description 5
- 238000011160 research Methods 0.000 description 5
- 238000012549 training Methods 0.000 description 5
- 230000003044 adaptive effect Effects 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 238000013507 mapping Methods 0.000 description 4
- 238000003908 quality control method Methods 0.000 description 4
- 238000000605 extraction Methods 0.000 description 3
- 230000005856 abnormality Effects 0.000 description 2
- 238000003062 neural network model Methods 0.000 description 2
- GNFTZDOKVXKIBK-UHFFFAOYSA-N 3-(2-methoxyethoxy)benzohydrazide Chemical compound COCCOC1=CC=CC(C(=O)NN)=C1 GNFTZDOKVXKIBK-UHFFFAOYSA-N 0.000 description 1
- 238000012935 Averaging Methods 0.000 description 1
- 230000001133 acceleration Effects 0.000 description 1
- 230000000739 chaotic effect Effects 0.000 description 1
- 230000001186 cumulative effect Effects 0.000 description 1
- 238000013481 data capture Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000005315 distribution function Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000014759 maintenance of location Effects 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 230000033764 rhythmic process Effects 0.000 description 1
- 238000010187 selection method Methods 0.000 description 1
- 230000009897 systematic effect Effects 0.000 description 1
- CFMYXEVWODSLAX-QOZOJKKESA-N tetrodotoxin Chemical compound O([C@@]([C@H]1O)(O)O[C@H]2[C@@]3(O)CO)[C@H]3[C@@H](O)[C@]11[C@H]2[C@@H](O)N=C(N)N1 CFMYXEVWODSLAX-QOZOJKKESA-N 0.000 description 1
- 229950010357 tetrodotoxin Drugs 0.000 description 1
- CFMYXEVWODSLAX-UHFFFAOYSA-N tetrodotoxin Natural products C12C(O)NC(=N)NC2(C2O)C(O)C3C(CO)(O)C1OC2(O)O3 CFMYXEVWODSLAX-UHFFFAOYSA-N 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 238000000844 transformation Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
- G06Q10/06395—Quality analysis or management
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/04—Manufacturing
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Human Resources & Organizations (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Strategic Management (AREA)
- Economics (AREA)
- Software Systems (AREA)
- Educational Administration (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Development Economics (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- Entrepreneurship & Innovation (AREA)
- General Business, Economics & Management (AREA)
- Life Sciences & Earth Sciences (AREA)
- Marketing (AREA)
- Tourism & Hospitality (AREA)
- Quality & Reliability (AREA)
- Operations Research (AREA)
- Game Theory and Decision Science (AREA)
- Manufacturing & Machinery (AREA)
- Primary Health Care (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a quality trend prediction method based on self-adaptive feature selection and improved thinking evolutionary algorithm, which mainly comprises three modules: the system comprises a characteristic self-adaptive processing module, a data fusion module and a quality trend prediction module. The method mainly comprises the following steps: (1) designing corresponding parameters to generate data for building the model; (2) Establishing a characteristic self-adaptive selection module by applying an error influence degree algorithm; (3) Establishing a data fusion module by applying a KPCA data fusion method; (4) A quality trend prediction module is established by optimizing a multi-layer perceptron (MLPNN) network by using an improved thinking evolution algorithm. By establishing the method, the method can be implemented in the field of quality trend prediction, can adaptively select different characteristics according to different types of data to predict, and adopts data fusion and algorithm improvement to improve the accuracy of product quality trend prediction, and timely adopts a proper mode to correct.
Description
Technical Field
The invention belongs to the field of intelligent manufacturing and intelligent quality monitoring, and particularly relates to a quality trend prediction method based on self-adaptive feature selection and improved thinking evolution algorithm.
Background
The method is widely applied in the fields of aerospace, aviation, ships, automobiles and the like, and because the quality data precision required by most of products in the fields is relatively high, the products can be affected by various factors such as people, machines, materials, methods, rings, measurement and the like in the production process, and meanwhile, the quality data often has the characteristics of timeliness, nonlinearity, correlation, dynamics and the like, so that the products can often generate larger quality problems, the trend can not be predicted in time, and the modification measures are proposed, so that the service performance and quality of the products are seriously affected.
Conventionally, control charts have been widely used in production processes as an aid to quality control and prediction. The upper and lower limits of the control chart control quality data capture fluctuation and abnormality of product quality, and along with the development of technology and the acceleration of production rhythm, the quality fluctuation is judged only by the upper and lower limits of the control chart, so that a large quality problem often occurs, and the quality control method cannot be suitable for modern means such as modern processing data acquisition and the like to control quality. At present, many scholars start to research the modes of the control chart to pursue the precise control of quality, but most of the research modes are few, do not research the mixed mode or cannot adapt to the dynamic change of the data, cannot perform the self-adaptive precise control, identification and prediction on the data, and mostly adopt offline identification, so that the degree of intelligence is not enough.
A quality trend prediction method based on self-adaptive feature selection and improved thinking evolution algorithm is provided for realizing intelligent quality control and improving the quality of products. The invention can realize intelligent quality prediction and control, and timely provides measures to correct and improve the product quality.
Disclosure of Invention
The purpose of the invention is that: aiming at the characteristics of dynamic property, time-varying property and the like of quality data, a quality trend prediction method based on self-adaptive feature selection and improved thinking evolution algorithm is provided based on various modes of a control diagram to realize multi-mode recognition, the quality trend of an online product in the dynamic production process is predicted according to the time-varying self-adaptive feature selection of the quality data, and meanwhile, the improved thinking evolution algorithm is also applied to optimize the recognition precision of the MLPNN network to improve the quality trend. The problems of few intelligent recognition modes, low precision, incapability of self-adaptive change and insufficient quality control strength of the existing quality trend are solved, and the yield is improved to achieve intelligent production and intelligent trend prediction.
The invention provides a quality trend prediction method based on self-adaptive feature selection and improved thinking evolution algorithm, which solves the problems, and builds a trend prediction model so as to be capable of predicting abnormal states. In order to achieve the above purpose, the present invention adopts the following technical scheme:
a quality trend prediction method based on self-adaptive feature selection and improved thinking evolutionary algorithm comprises the following steps:
step 1: generating a model and establishing required data.
The data for generating 9 modes are normal mode (NOR), periodic mode (CYC), system mode (SYS), hierarchical mode (STR), upward trend mode (IT), downward trend mode (DT), upward step mode (US), downward step mode (DS), and mixed Mode (MIX), respectively.
Step 2: and establishing a characteristic self-adaptive processing module.
In the quality trend prediction process, the general common method directly applies data to predict trend with low accuracy, and the establishment of the step can reasonably select the characteristics and further improve the intelligent degree. The module is built by two steps:
the first step: the feature extraction model is established, and the extracted data features comprise the following statistical features and shape features according to the research shows that the data features are more convincing; wherein, the statistical characteristics of the quality data include: MEAN, VS, STD, SKEW, KURT, A; the shape characteristics of the quality data include: SL, NC1, NC2, APML, APLS, AASL, ACLPI, SRANGE, SB, PSMLSC, REAE, ABDPE.
The following features are extracted from the quality data in the observation window, and the meanings of the symbols are as follows:
MEAE: the mean value of the quality data;
VS: mean square value of the quality data;
STD: standard deviation of quality data;
skuw: a bias coefficient of the quality data;
KURT: kurtosis coefficient of the mass data;
a: autocorrelation coefficients of the quality data;
SL: the slope of a least squares regression line fitted to the quality data;
NC1: the number of crossing points of the curve formed by the quality data and the line formed by the average value;
NC2: the number of cross points of the curve formed by the quality data and the least square regression line;
APML: the area between the curve formed by the mass data and the average line;
APLS: the area between the curve formed by the quality data and the least square regression line;
AASL: the curve formed by the mass data is divided into four areas, and the average value of the slope of the curve formed by the midpoint combination connecting line of each area
ACLPI: the ratio of APML to standard deviation of the quality data;
SRANGE: the curve formed by the mass data is divided into four areas, and the difference between the maximum value and the minimum value of the slope formed by the midpoint connecting line in each area;
SB: slope identifiers of the quality data least squares regression line;
PSMLSC: an average value of the intersection point sum of the quality data and the central line and the least square regression line;
read: the ratio of the error of the MSE of the quality data to the average value of the MSE of the data divided into four areas;
ABDPE: when the curve formed by the mass data is divided into two areas, the absolute value of the difference between the slope of the least square regression line of the curve formed by the whole mass data and the average value of the slopes of the least square regression line of the two areas.
And a second step of: and establishing a self-adaptive feature selection model, and establishing a method for introducing error to influence calculation by the model. If new data exists, the module can be applied to select proper characteristics according to the data characteristics. Based on different kinds of quality data, different features are selected according to the importance degree of the features.
The method for establishing the model is as follows:
a) Setting N quality data features, and preprocessing the quality data features;
b) Establishing an initial MLPNN neural network, and training by applying the processed quality data characteristics;
c) If new data is input, preprocessing new quality data characteristics, respectively increasing the value of each quality data characteristic by 10% and reducing the value by 10% to generate 2N groups of data, respectively taking each group of data into the MLPNN for identification to obtain 2N error values, and then averaging the errors obtained by increasing the value by 10% and reducing the value by 10% corresponding to each quality data characteristic to obtain N error values;
d) The first 85% of features selected from the order of the errors from large to small are defined as the features with higher degrees of influence on the quality data.
e) And (5) completing the self-adaptive quality data feature selection model.
Step 3: and a data characteristic fusion module.
In order to more accurately predict the trend, the original data is fused with the adaptively selected data. If the quality trend of the original data is predicted directly with the selected characteristics, the input data is too huge, the calculation complexity of the model is greatly increased, the data dimension reduction method is divided into linear data dimension reduction and nonlinear data dimension reduction, the module adopts a KPCA data dimension reduction method to fuse the original data with the characteristic data, and the method for establishing the module comprises the following steps:
a) The combined data is standardized and centralized;
b) Constructing a kernel function of the combined data, mapping the data into a high dimension and calculating a kernel matrix;
c) The calculated eigenvalue and the selected eigenvector;
d) Performing data dimension reduction and fusion;
step 4: and establishing a quality trend prediction module.
And establishing a 3-layer perceptron MLPNN neural network model based on the data after the self-adaptive feature fusion, and optimizing the weight and the threshold of the MLPNN neural network by using an improved thinking evolution algorithm.
The general thinking evolution algorithm mainly adopts an iterative optimization learning mode, all individuals in the evolution process are called groups, and one group is divided into a plurality of subgroups. The subgroups include winning subgroups and temporary subgroups. During thinking evolution, the winning subgroup and the temporary subgroup are randomly generated by taking the optimal particles as the center in the flying evolution process without any limitation, and the information degree contained among the particles in the subgroup cannot be judged, so that particles which have no meaning on the evolution are generated. Mutual information theory is introduced here to determine the goodness of subgroup evolution. When the information degree contained by the particles and the central particles generated in the subgroup is more than 85%, the particles are considered as invalid particles, meanwhile, if the score of a certain particle is more than that of the central particle, the particles are reserved, otherwise, the particles are regenerated.
The method for establishing the improved thinking evolutionary algorithm comprises the following steps:
a) Initializing population generation, and generating a population in a space;
b) The higher scoring particles in the initialized population are selected as the winning subgroup center, temporary subgroup center, and the generating subgroup, respectively.
c) Introducing an information judgment operator: calculating mutual information degrees between particles of each winning subgroup, temporary subgroup and central particles of the own subgroup respectively, if the mutual information degree of individual particles and the central particles is more than 85%, the particles are considered to be similar particles, and meanwhile, if the individual particles are higher than the central particles, the particles are reserved, otherwise, the particles are released;
d) Convergence operator: individual particle scores in all subgroups are calculated, and winners are selected as centers to regenerate the subgroups. And c, if the information degree of the individual particles and the central particles in the subgroup is proper, continuing the next operation, otherwise, regenerating the particles and performing the operation of the step c.
e) Judging whether each subgroup is mature or not, if so, continuing the next operation, otherwise, continuing the operation of the step d.
f) Dissimilating operation operators: all mature winning subgroups and temporary subgroups are communicated with information, and temporary subgroups with higher scores will replace temporary subgroups with lower scores.
g) And c, judging that if no score exceeds the mature winning subgroup in the mature temporary subgroup, jumping out of the circulation, otherwise, repeating the operations of the steps c-g.
Meanwhile, in the operation, when a new subgroup is generated by using a new center particle in the step d, an entropy change theory is introduced to increase the chaos degree of particle generation and increase the entropy change inertia coefficient. The method has the advantages that the requirement range is larger in the early search period and a certain relation is kept between the requirement range and the information of the center particles, the convergence is faster in the later search period, and the searching capability of excellent particles can be improved by introducing the inertia coefficient.
By analyzing abnormal conditions in the production process, a method for adaptively selecting proper features is formulated according to the characteristics of quality data, the adaptively selected features are fused with original data to reduce dimensionality and increase recognition accuracy, and an information judgment operator and an entropy increase theory improved thinking evolution algorithm are introduced to improve the accuracy and searching capability of the algorithm, so that the method has the advantages that: the proper characteristics can be selected in a self-adaptive mode according to the characteristics of the quality data, so that the recognition accuracy is ensured; the use of the data fusion method can enhance the quality trend prediction performance and ensure that the recognizer has good training efficiency; the improved algorithm can ensure good fault tolerance and stronger classification capability in classification.
Drawings
The invention will be further described with reference to the accompanying drawings and examples
FIG. 1 is a flow chart of the present method;
FIG. 2 is an image of 9 modes; (a) a normal mode; (b) a mixed mode; (c) a periodic pattern; (e) a system mode; (f) a hierarchical mode; (g) upward trend; (h) a downward trend; (i) an upward step trend; (j) lower step mode;
FIG. 3 is a distribution diagram of 9 pattern correspondence features;
FIG. 4 is a schematic illustration of feature fusion
FIG. 5 is a flow chart of an improved thought evolution algorithm
FIG. 6 is a schematic diagram of a particle reduction of the thought evolution algorithm
Detailed Description
According to the invention, the dynamic property and time-varying property of quality data during production are considered to introduce an adaptive feature selection method, the dynamic data and the adaptive feature are fused for accurately identifying the application of a KPCA data fusion method, and an accurate identification model is established by finally applying an MLPNN network and combining an improved thinking evolution algorithm. The invention will be further described with reference to the drawings and specific embodiments thereof:
an overall flow chart of the implementation of the method is shown in fig. 1, and the following is a specific expansion for each module.
Step 1: generating a model and establishing required data.
The patterns that may occur during the production process are analyzed. The fluctuation of the product quality appears in a certain rule and affects the subsequent production under the influence of the actual production. Designing different parameters to generate different types of trend graphs as shown in fig. 2 is a graph of 9 modes: the data of the normal mode, the mixed mode, the periodic mode, the layering mode, the system mode, the upward trend mode, the downward trend mode, the upward step mode and the downward step mode are specifically as follows:
the simulation formula: y=μ+r (t) +d (t)
Wherein: mu is the average value of quality data, R (t) is the normal distribution random deviation at the moment t, d (t) is the deviation caused by abnormal factors in the production process, and the following formula and parameter description of each mode during production are as follows:
(1) Normal mode of production (NOR):
y＝μ+r(t)×σ+d(t)
wherein: mu is the average value of quality data, r (t) is a standard normal random distribution function, d (t) is fluctuation caused by abnormal factors, d (t) =tetrodotoxin in a normal mode, and the recommended sigma value is 0.05 at the minimum and 0.5 at the maximum;
(2) Mixed mode of production process (MIX):
y＝μ+r(t)×σ+(-1) W ×m×σ
wherein: d (t) is fluctuation caused by an abnormal factor, and suggested sigma takes a minimum value of 0.05 and a maximum value of 0.5; w is a disturbance factor 0 or 1 in the production process, and m is a disturbance amplitude suggestion 1.5,2.5 in the production process.
(3) Periodic pattern (CYC) of the production process:
y＝μ+r(t)×σ+a×sin(2πt/T)×σ
wherein: d (T) is fluctuation caused by an abnormal factor, and d (T) =a×sin (2pi T/T) ×σ in a periodic mode, wherein the value of σ is recommended to be 0.05 at the minimum and 0.5 at the maximum; suggesting a minimum value range [1,1.5] of the amplitude value a of the disturbance of the periodic production process, wherein the maximum value range is [2.5,3]; it is recommended that the period T of the process disturbance has a minimum value in the range of [4,8], and the period Tmax has a minimum value in the range of [10,16].
(4) Layering mode of production process (STR):
y＝μ+r(t)×σ×K
wherein: d (t) is fluctuation caused by an abnormal factor, and suggested sigma takes a minimum value of 0.05 and a maximum value of 0.5; the minimum value range of the disturbance proportion relation K of the production process data is recommended to be [0.1,0.3], and the maximum value range of K is recommended to be [0.4,0.6].
(5) Systematic mode of production process (SYS):
y＝μ+r(t)×σ+d×(-1) t ×σ
wherein: d (t) is fluctuation caused by an abnormality factor, and d (t) =d× (-1) in the system mode t X sigma, recommended sigma value is minimum of 0.05 and maximum of 0.5; suggesting that the deviation degree dmin of the quality data is 1; d has a maximum value of 3.
(6) Up-down trend pattern (IT/DT) of the production process:
y＝μ+r(t)×σ±g×t×σ
wherein: d (t) is fluctuation caused by abnormal factors, and d (t) = ±g×t×σ in a trend mode, wherein the suggested value of σ is 0.05 at minimum and 0.5 at maximum; the value range of the slope g of the recommended quality data is 0.05, 0.1;
(7) Up and down step mode (US/DS) of production process:
y＝μ+r(t)×σ±b×s×σ
wherein: d (t) is fluctuation caused by an abnormal factor, and d (t) = ±b×s×σ in a step mode, wherein the value of σ is recommended to be 0.05 at minimum and 0.5 at maximum; b=0 when t < P, b=1 when t > P; suggesting that the random step position Pmin value of the quality data is [4,9], and the Pmax value range is [13,19]; the range of the minimum value of the step amplitude value s is recommended to be [0.5,1.5], and the range of the maximum value s is recommended to be [2.5,3.5].
Step 2: feature adaptive processing module establishment
In the quality trend prediction process, the general common method directly applies data to predict trend, and researches show that reasonable quality data feature use greatly improves prediction accuracy. The quality data features are extracted, and the specific feature distribution is shown in fig. 3. The establishment of the characteristic self-adaptive processing module comprises two steps:
the first step: establishing a feature extraction model, wherein the step of extracting statistical features of quality data comprises the following steps: MEAN, VS, STD, SKEW, KURT, A; the shape characteristics of the quality data include: SL, NC1, NC2, APML, APLS, AASL, ACLPI, SRANGE, SB, PSMLSC, REAE, ABDPE. Let the quality data be y= (Y) 1 ,y 2 ,y 3 ....y P ) (i=1, 2, 3P), the feature extraction method comprises the following steps:
mean value formula of quality data:
the mean square value formula of the quality data:
standard deviation of quality data
The bias coefficient formula of the quality data:
the kurtosis coefficient formula of the mass data:
autocorrelation coefficients of quality data:
the slope of the least squares regression line of the curve formed by the mass data:
t i the distance from the origin to the detected mass data point at the ith time,is the average of the P detection mass data points from the origin. y is i Is the ith quality data.
The number of cross points of the curve formed by the mass data and the curve formed by the average value: if (y) i -MEAN)(y i+1 -MEAN)<0, NC1 is added with 1;
the number of cross points of the curve formed by the quality data and the least square regression line: if (y) i -L(y i ))(y i+1 -L(y i+1 ))<0, NC2 is added with 1;
area between the curve formed by the mass data and the average line: APML;
area of the curve formed by the mass data and the least squares regression line: APLS;
dividing the mass data into four sections of areas, and obtaining a slope average value by connecting points in two areas: wherein S is jk The slope of the mass data region is composed of the jth midpoint and the kth midpoint. The midpoint coordinates of each region are +.>
Ratio of area of mass data to center line composition to standard deviation of mass data:
the quality data is divided into four areas, and the difference between the maximum value and the minimum value of the slope formed by the midpoint connecting line in each area is: srange=max (S jk )-min(S jk )；(j＝123；k＝234；J<k)
The slope identifier of the least squares regression line of the curve formed by the quality data: SB is 1 if the least square slope of the curve formed by the mass data is greater than 0, otherwise SB is 0;
intersection point of curve formed by quality data and central line, and least square backAverage value of sum of intersection points of the return lines:
error ratio of MSE of quality data to MSE average of data divided into four regions:
when the mass data is divided into two areas, the absolute value of the difference between the slope of the integral least squares regression line and the average value of the slopes of the least squares regression line of the two areas:wherein B is the slope of the least squares regression line of the whole curve formed by the mass data, ++>Is the average of the slopes of the least squares regression line for the two regions.
And a second step of: and establishing a self-adaptive feature selection model, and establishing a method for introducing error to influence calculation by the model. In the production process, the important features implied by different kinds of quality data are different, and the important features are selected based on different data. The specific method for establishing the model comprises the following steps:
a) Preprocessing the quality data characteristics;
let feature matrix f= (F) ab ) (a=1, 2, 3..h; b=1, 2, 3..n), where there are H samples, N classes of features, the quality data features were normalized:wherein: />Is of class b qualityMean value of data features, s b Standard deviation +.>
b) Establishing a 3-layer perceptron neural network with hidden layers asThe number of the nodes is m, the number of the input nodes is n, the number of the output nodes is n, and training data is selected from the quality data characteristics to carry out MLPNN neural network training;
c) If the quality data feature f is newly input 1 ,f 2 ,f 3 ....f N ]Preprocessing the quality data characteristics to obtain F' = (F) b ') (b=1, 2, 3..n), and adding 10% and subtracting 10% of each quality data characteristic value, respectively, together generate 2N sets of data [ f 1 * ,f 2 * ,f 3 * ....f 2N * ]Wherein f 1 * -f N * Adding 10% data for each quality data characteristic value, f N+1 * -f 2N * In order to reduce the numerical value of each quality data characteristic by 10% of data, two groups of quality data characteristics are respectively brought into MLPNN to be identified to obtain 2N trend prediction errors respectively as [ E ] 1 U ,E 2 U ,E 3 U ....E N U ]、[E 1 D ,E 2 D ,E 3 D ....E N D ]Then, the trend prediction errors obtained by correspondingly increasing each quality data characteristic by 10 percent and reducing the quality data characteristic by 10 percent are averaged to obtain an average value E i ＝(E i U +E i D ) 2 (i=123.). N) obtaining N Individual trend prediction error value [ E ] 1 ,E 2 ,E 3 ....E N ]；
d) Will [ E ] 1 ,E 2 ,E 3 ....E N ]The first 85% of features selected from the order of the error are defined as features with higher degrees of influence on the quality data.
e) And (3) completing the self-adaptive quality data feature selection to obtain L (L < N) features, and completing the establishment of the model. Preparation is made for the subsequent feature fusion.
Step 3: data feature fusion module
And establishing a data characteristic fusion module, and fusing the original data with the self-adaptive selected data in order to more accurately predict the trend. The module adopts a KPCA data dimension reduction method to fuse the original data with the characteristic data as shown in figure 4, and the specific implementation method comprises the following steps:
a) The combined data is standardized and centralized;
let data matrix f= (F) ab ) (a=1, 2, 3..h; b=1, 2,3. (l+p)), where there are H samples, each sample having l+p data, the combined data were normalized:wherein: />Mean value of b-th dimension data in combined data, s b Standard deviation of b-th dimensional data after combining quality data and quality data characteristics +.> Obtaining F a ＝[f a1 ″,f a2 ″,f a3 ″....f a(L+P) ″]，F″＝[F 1 ,F 2 ....F H ]。
b) Constructing a combined data kernel function, mapping the processed data into a high dimension, and calculating a kernel matrix of the combined data;
the load factor of the combined data is selected according to the eigenvalue lambda and eigenvector W of the combined data, and fusion data is generated, wherein the nonlinear mapping phi is introduced, so that the problem can be converted into phi (F') T W＝λW。
Mapping function incorporating combined dataVector matrix w= (W) of the combined data is obtained 1 ,w 2 ,w 3 ....w H )，w a For a base vector of combined data, the relationship of the mapped data may be expressed as
Is available in the form of
Order theTo obtain alpha= [ alpha ] 1 ,α 2 ....α H ]
Finally, phi (F') T Φ(F″)α＝λΦ(F″)α
The core skill equation is introduced here with both sides multiplied by phi (F') T The method comprises the following steps:
c) Calculating the characteristic value of the combined data and selecting a characteristic vector;
kernel functions incorporating combined data and performing kernel-to-skill transformations K (F a1 ,F a2 )＝Φ(F a1 )Φ(F a2 ) T Wherein (a 1 e (1, H), a2 e (1, H)); at this time, the first g principal components can be defined to represent the original data matrix when the cumulative contribution rate of the eigenvalues is greater than 85%, and the load factor matrix after the matrix selection with K being H×H at this time is A= [ alpha ] 1 ,α 2 ....α g ]。
d) Performing data dimension reduction;
multiplying the combined data by the selected load vector to obtain fused data X * The =k×a is:
step 4: establishing a quality trend prediction module
The data after the self-adaptive feature fusion is used for establishing a 3-layer perceptron MLPNN neural network model, an improved thinking evolution algorithm is used for optimizing the weight and the threshold of the MLPNN neural network, and the training data adopts the data fused in the 3 rd step.
The general thinking evolution algorithm mainly adopts an iterative optimization learning mode, all individuals in the evolution process are called groups, and one group is divided into a plurality of subgroups. Subgroups include winning and temporary groups. During thinking and evolution, the winning particle swarm and the temporary particle swarm are randomly generated by taking the optimal particles as the center in the flying and evolution process, and the information degree contained among the particles in the subgroup cannot be judged, so that a plurality of particles which have no meaning on the evolution are likely to be generated, and the particles have no meaning on the evolution. Mutual information theory is introduced here to determine how good a subgroup evolves. When the information degree contained by the particles and the leading particles generated in the subgroup is greater than 85%, the particles are considered as invalid particles, meanwhile, if the score of a certain particle is greater than that of the leading particle, the particles are reserved, otherwise, the leading particle is taken as a center to generate the particles, as shown in fig. 5, which is a step of a specific operation, and fig. 6 is a schematic diagram of a particle evolution process.
The method for establishing the improved thinking evolutionary algorithm comprises the following steps:
a) Initializing population generation, generating a population with T particles in space, and calculating the score of each particle according to an adaptive function;
b) In the initialized population, A particles with higher scores are selected as the centers of winning subgroups, and B particles with higher scores are selected as the centers of temporary subgroups. Setting the size of each subgroup to be T * Wherein T is * ＝T/(A+B)。
c) Introducing an information judgment operator: calculating the center of each particle of the winning subgroup respectively with the particle length L Particle centers of temporary subgroups and->Mutual information between->Is->Wherein (i= 1,2,3. T * ,l＝1 2 3.....L′ t 1 ＝1,2,3,...A，t 2 =1, 2, 3..b.) if the individual particles of both subgroups have a mutual information degree of greater than 85% with the central particle, the particles are considered to be similar particles, while rules are formulated if the particle score is greater than the central particle retention, otherwise release;
taking the calculation of the mutual information degree of the winning subgroup as an example:
first, the information entropy formula for calculating the winning subgroup is as follows:
wherein the method comprises the steps ofIndicating individual particles->Is generally used logarithmically here with a base of 2, e or 10.
The degree of mutual information for the center particle and the individual particles can be expressed as:
the same applies to the temporary subgroup G as well as to the above formula.
d) Convergence operator: and generating A winning subgroups and B temporary subgroups by taking the selected particles as winning centers and subjecting the temporary centers to normal distribution, calculating individual particle scores in all the subgroups, and selecting winners as centers to regenerate the subgroups. And c, if the information degree of the individual particles and the central particles in the subgroup is proper, continuing the next operation, otherwise, regenerating the particles and performing the operation of the step c.
e) Judging whether each subgroup is mature or not, if so, continuing the next operation, otherwise, continuing the operation of the step d.
f) Dissimilating operation operators: communicating information between all mature winning subgroups and temporary winning subgroups, winning subgroup M 1 ,M 2 ...M A Temporary subgroup G 1 ,G 2 ,G 3 ...G B Competing, if there is a mature temporary subgroup score greater than a mature winning subgroup score, then release of winning particles is replaced by and the temporary subgroup is replenished.
g) And c, judging that if the score in the mature temporary subgroup does not exceed the score of the mature winning subgroup, jumping out of the cycle, otherwise, repeating the operation of the step c-the step g.
Meanwhile, in the above operation, when a subgroup is generated centering on the optimal particles in the step d, generation of new individual particles is required. The new individual particle generation introduces an entropy change theory to increase the chaos degree of particle generation, and the entropy change inertia coefficient is added into a particle generation formula. The method has the advantages that the requirement range is larger in the early search period and a certain relation is kept between the requirement range and the information of the center particles, the convergence is faster in the later search period, and the searching capability of excellent particles can be improved by introducing the inertia coefficient.
The method for generating new individual particles proceeds in the following manner:
a) To expand the range search, an initial particle matrix R is generated according to a formula D×L′ Wherein the number of particles is D and the length of each particle is L'. The formula for particle generation is: r is R d ＝m c +×θ×(2×Z 1×L′ -1), wherein θ is 0.5 to 1, Z 1×L′ Is 1 XL' normal distribution 0-1 value matrix, m c Is a center particle.
b) For R D×L′ The matrix carries out information entropy calculation of the particles and carries out chaos iteration of the particles. Setting the iteration number as K max And k is the value of the number of iterations, and finally the optimal particle is selected. Converting a particle matrix into S 1×(D×L′) The information entropy calculation formula of the matrix of the whole particle is as follows:applying a formula to obtain a variable weight:
c) Changing the formula to R d+1 ＝wR d +(1-w)×θ×(2×Z 1×L′ -1) cycle K max After a second time, a new R is obtained D×L′ Meanwhile, the high score of each particle can be reserved in the circulation process, and the particle with the highest score in the matrix is finally selected as a new particle after the circulation is finished.
The process is a chaotic search process in which a certain particle is taken as the center and is in the range, so that the particle search is prevented from exceeding the fitting search, the range is expanded, and finally the particle is gradually close to the optimal particle, and the search precision around the center particle is improved.
Claims (5)
1. A quality prediction and control method is characterized in that: modeling of the method comprises the following four steps:
step 1: generating product quality model building required data, and simulating quality data of the product; the modes of the quality data of the simulation product comprise a normal mode, a periodic mode, a mixed mode, a system mode, an ascending trend mode, a descending trend mode, an upward step mode and a downward step mode;
step 2: and (3) establishing a characteristic self-adaptive processing module, preprocessing according to the quality data of the product in the step (1), wherein the characteristic self-adaptive processing module is established by two steps: extracting product quality data characteristics and establishing a self-adaptive characteristic selection model by applying an initialization MLPNN network;
step 3: establishing a data feature fusion module, and realizing feature fusion and data dimension reduction by applying a KPCA method, thereby simplifying a subsequent product quality trend prediction module;
step 4: establishing a product quality trend prediction module, optimizing an MLPNN neural network by using an improved thinking evolution algorithm, wherein the optimization target of the model is prediction accuracy, and introducing an entropy change theory by adding a mutual information judgment operator to make the algorithm obtain a prediction model;
the method for establishing the improved thinking evolutionary algorithm comprises the following steps:
a) Initializing population generation, and generating a population in a space;
b) Selecting particles with higher scores in the initialized population as a winning subgroup center and a temporary subgroup center respectively and generating subgroups;
c) Introducing an information judgment operator: calculating mutual information degrees between particles of each winning subgroup, temporary subgroup and central particles of the own subgroup respectively, if the mutual information degree of individual particles and the central particles is more than 85%, the particles are considered to be similar particles, and meanwhile, if the individual particles are higher than the central particles, the particles are reserved, otherwise, the particles are released;
d) Convergence operator: calculating individual particle scores in all subgroups, and selecting winners as centers to regenerate the subgroups; c, if the information degree of the individual particles and the central particles in the subgroup is proper, continuing the next operation, otherwise, regenerating the particles and performing the operation of the step c;
e) Judging whether each subgroup is mature or not, if so, continuing the next operation, otherwise, continuing the operation of the step d;
f) Dissimilating operation operators: carrying out information communication on all mature winning subgroups and temporary subgroups, wherein the temporary subgroup with higher score replaces the temporary subgroup with lower score;
g) And c, judging that if no score exceeds the mature winning subgroup in the mature temporary subgroup, jumping out of the circulation, otherwise, repeating the operations of the steps c-g.
2. The quality prediction and control method of claim 1, wherein: the data of the 9 modes generated in the step 1 are respectively: normal mode NOR, periodic mode CYC, system mode SYS, hierarchical mode STR, upward trend mode IT, downward trend mode DT, upward step mode US, downward step mode DS, mixed mode MIX.
3. The quality prediction and control method of claim 1, wherein: the statistical features of the quality data adopted in the first step in the step 2 include: MEAN, VS, STD, SKEW, KURT, A; the shape characteristics of the quality data include: SL, NC1, NC2, APML, APLS, AASL, ACLPI, SRANGE, SB, PSMLSC, REAE, ABDPE; and secondly, adopting an initialized MLPNN network to apply error influence degree algorithm to adaptively select the characteristics.
4. The quality prediction and control method of claim 1, wherein: and 3, fusing the self-adaptively selected characteristics with the original product quality data by using a KPCA method.
5. The quality prediction and control method of claim 1, wherein: and 4, designing a thinking evolutionary algorithm improved by the mutual information judgment operator, and improving a particle generation mode of the thinking evolutionary algorithm by applying an entropy change theory.
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010405648.2A CN111598435B (en) | 2020-05-14 | 2020-05-14 | Quality trend prediction method based on self-adaptive feature selection and improved thinking evolutionary algorithm |
| PCT/CN2020/127970 WO2021227406A1 (en) | 2020-05-14 | 2020-11-11 | Quality trend prediction method based on adaptive feature selection and improved mind evolutionary algorithim |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010405648.2A CN111598435B (en) | 2020-05-14 | 2020-05-14 | Quality trend prediction method based on self-adaptive feature selection and improved thinking evolutionary algorithm |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111598435A CN111598435A (en) | 2020-08-28 |
| CN111598435B true CN111598435B (en) | 2023-08-04 |
Family
ID=72190810
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010405648.2A Active CN111598435B (en) | 2020-05-14 | 2020-05-14 | Quality trend prediction method based on self-adaptive feature selection and improved thinking evolutionary algorithm |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN111598435B (en) |
| WO (1) | WO2021227406A1 (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111598435B (en) * | 2020-05-14 | 2023-08-04 | 北京工业大学 | Quality trend prediction method based on self-adaptive feature selection and improved thinking evolutionary algorithm |
| CN114496209B (en) * | 2022-02-18 | 2022-09-27 | 青岛市中心血站 | Intelligent decision-making method and system for blood donation |
| CN114997070B (en) * | 2022-07-15 | 2022-11-11 | 合肥中科迪宏自动化有限公司 | Training method of control chart pattern recognition model and control chart pattern recognition method |
| CN117310118B (en) * | 2023-11-28 | 2024-03-08 | 济南中安数码科技有限公司 | Visual monitoring method for groundwater pollution |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105139093A (en) * | 2015-09-07 | 2015-12-09 | 河海大学 | Method for forecasting flood based on Boosting algorithm and support vector machine |
| CN106769030A (en) * | 2016-11-10 | 2017-05-31 | 浙江工业大学 | A kind of bearing state tracking and Forecasting Methodology based on MEA BP neural network algorithms |
| CN110047015A (en) * | 2019-04-22 | 2019-07-23 | 水利部信息中心 | A kind of water total amount prediction technique merging KPCA and thinking Optimized BP Neural Network |
| CN111027745A (en) * | 2019-11-08 | 2020-04-17 | 广东财经大学 | Stock index prediction method based on self-adaptive feature extraction |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111598435B (en) * | 2020-05-14 | 2023-08-04 | 北京工业大学 | Quality trend prediction method based on self-adaptive feature selection and improved thinking evolutionary algorithm |
-
2020
- 2020-05-14 CN CN202010405648.2A patent/CN111598435B/en active Active
- 2020-11-11 WO PCT/CN2020/127970 patent/WO2021227406A1/en active Application Filing
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105139093A (en) * | 2015-09-07 | 2015-12-09 | 河海大学 | Method for forecasting flood based on Boosting algorithm and support vector machine |
| CN106769030A (en) * | 2016-11-10 | 2017-05-31 | 浙江工业大学 | A kind of bearing state tracking and Forecasting Methodology based on MEA BP neural network algorithms |
| CN110047015A (en) * | 2019-04-22 | 2019-07-23 | 水利部信息中心 | A kind of water total amount prediction technique merging KPCA and thinking Optimized BP Neural Network |
| CN111027745A (en) * | 2019-11-08 | 2020-04-17 | 广东财经大学 | Stock index prediction method based on self-adaptive feature extraction |
Non-Patent Citations (1)
| Title |
|---|
| 初红艳 ; 李鹏 ; 蔡力钢 ; 李风光 ; .基于MC方法和BP网络的印刷质量控制图模式识别研究.北京工业大学学报.(第06期),第816页数第1节-第820页第4节. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111598435A (en) | 2020-08-28 |
| WO2021227406A1 (en) | 2021-11-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111598435B (en) | Quality trend prediction method based on self-adaptive feature selection and improved thinking evolutionary algorithm | |
| CN109141847B (en) | Aircraft system fault diagnosis method based on MSCNN deep learning | |
| CN113609596B (en) | Aircraft aerodynamic characteristic prediction method based on neural network | |
| CN108564136B (en) | A kind of airspace operation Situation Assessment classification method based on fuzzy reasoning | |
| CN112838946B (en) | Method for constructing intelligent sensing and early warning model based on communication network faults | |
| CN109389171B (en) | Medical image classification method based on multi-granularity convolution noise reduction automatic encoder technology | |
| CN110909762B (en) | Robot posture recognition method and device based on multi-sensor fusion | |
| CN111311702B (en) | Image generation and identification module and method based on BlockGAN | |
| CN102222240B (en) | DSmT (Dezert-Smarandache Theory)-based image target multi-characteristic fusion recognition method | |
| CN112597702B (en) | Pneumatic modeling generation type confrontation network model training method based on radial basis function | |
| CN111680875A (en) | Unmanned aerial vehicle state risk fuzzy comprehensive evaluation method based on probability baseline model | |
| CN114170789A (en) | Intelligent network connected vehicle lane change decision modeling method based on space-time diagram neural network | |
| CN110851654A (en) | Industrial equipment fault detection and classification method based on tensor data dimension reduction | |
| CN116738339A (en) | Multi-classification deep learning recognition detection method for small-sample electric signals | |
| CN107590538B (en) | Danger source identification method based on online sequence learning machine | |
| CN114254915A (en) | Method for deciding and optimizing qualified state of full-flow processing quality of shaft parts | |
| CN117875201A (en) | Quantitative evaluation method for interstage separation scheme | |
| CN114118592A (en) | Short-term energy consumption prediction system for power consumption end of smart power grid | |
| CN109164794A (en) | Multivariable industrial process Fault Classification based on inclined F value SELM | |
| CN115206455B (en) | Deep neural network-based rare earth element component content prediction method and system | |
| CN110414079A (en) | One kind having causal inconsistent data processing method | |
| CN115146466A (en) | System failure probability calculation method under multi-failure mode based on multi-point and point-adding criterion | |
| CN115081483A (en) | Hydro-generator rotor fault diagnosis method based on feature selection and GWO-BP | |
| Liu et al. | Real-time Prediction Method of Remaining Useful Life Based on TinyML | |
| CN113807005A (en) | Bearing residual life prediction method based on improved FPA-DBN |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |