CN109145948A - A kind of injection molding machine putty method for detecting abnormality based on integrated study - Google Patents
A kind of injection molding machine putty method for detecting abnormality based on integrated study Download PDFInfo
- Publication number
- CN109145948A CN109145948A CN201810790797.8A CN201810790797A CN109145948A CN 109145948 A CN109145948 A CN 109145948A CN 201810790797 A CN201810790797 A CN 201810790797A CN 109145948 A CN109145948 A CN 109145948A
- Authority
- CN
- China
- Prior art keywords
- data
- data set
- putty
- classifier
- injection molding
- 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.)
- Withdrawn
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Injection Moulding Of Plastics Or The Like (AREA)
Abstract
The present invention relates to a kind of injection molding machine putty method for detecting abnormality based on integrated study, including raw data acquisition, data prediction, characteristics extraction, construct putty anomaly classification device, in the characteristics extraction stage, the temporal signatures of each module operating voltage of extraction place first from sample database, then these characteristic values input second level anomaly classification device is trained, three base classifiers are constructed first, the highest result of the frequency of occurrences is selected to combine original characteristic data set to be passed to second level strong classifier as newly-increased characteristic series using ballot method the result of three base classifiers, finally, the result of strong classifier is exactly the detected value of injection molding machine putty exception.The present invention can quickly and accurately have found the unusual condition of injection molding machine putty, the not high feature of original detection mode accuracy is overcome, the implementation steps of the invention is simple and convenient, can detect in time to injection molding machine putty, facilitate the normal operation for being molded machine equipment, extends the service life of equipment.
Description
Technical field
The present invention relates to the applications of machine learning algorithm and injection molding machine putty abnormality detection the relevant technologies, more particularly to one kind
Injection molding machine putty method for detecting abnormality based on integrated study.
Background technique
In recent years, the continuous development of Internet technology, machine learning algorithm start to be cured gradually in the application field of all trades and professions
Extensively, in industrial circle, machine learning related algorithm is applied among machine abnormality detection or failure modes;
Injection molding machine is using most commonly seen equipment in field of plastics processing, and in actual moving process, injection molding machine is often being fed
Occur the problem of putty in the process, thus promote the disqualification rate of product, putty can also cause injection molding machine itself once for a long time
Damage, therefore, how machine learning related algorithm to be applied among injection molding machine putty abnormality detection has realistic meaning very much.
Machine learning algorithm is broadly divided into four classes at present, respectively supervised learning, unsupervised learning, semi-supervised learning and strong
Chemistry practise, what is applied in fault detection is supervised learning algorithm comprising K- nearest neighbor algorithm, decision tree, naive Bayesian,
SVM algorithm and decision Tree algorithms, but be all at present that single detection supervised learning art is taken to research and develop for abnormality detection mode,
Cause the accuracy of final result low.
Summary of the invention
In view of the above-mentioned state of the art, technical problem to be solved by the present invention lies in provide one kind to promote inspection
Survey the injection molding machine putty method for detecting abnormality based on integrated study of result accuracy.
The technical scheme of the invention to solve the technical problem is: a kind of injection molding machine putty based on integrated study
Method for detecting abnormality, which is characterized in that method includes the following steps:
1) raw data acquisition
Module voltage collector is installed on each operational module of injection molding machine, and each work of timing acquiring within the set duration
The operating voltage made in module is stored as initial data into database;
2) data prediction
Data are successively used to operating voltage each group of data of each operational module stored in database within the assigned work time
Cleaning method, data normalization method pretreatment;
3) characteristics extraction
Pretreated data are obtained to step 2 and extract characteristic value, composition with temporal signatures (mean value, variance, standard deviation, in
Digit, peak value, valley) data set, and store into document data set;
4) data set positive and negative samples are generated
New data set in data file is divided into putty abnormal data set and works normally data set, and using synthesis minority class
Oversampling technique (SMOTE) increases the data set of putty abnormal conditions, the positive and negative samples that equilibrium data is concentrated;
5) putty anomaly classification device is constructed
Secondary classifier is constructed on the basis of step 4, the first order is three base classifiers, and the second level is a classifier, the
Three base classifiers of level-one are respectively Logistic classifier, SVM classifier and decision tree classifier, and second level classifier is
Logistic classifier.
Further, the electricity of each operational module of injection molding machine is periodically acquired in the step 1 in setting duration section
When pressing data, need to record the product specification of injection molding machine production in real time, once the period for putty exception occur is labeled as 1,
Remaining work normally is labeled as 0.
Further, the data cleaning method in the step 2 is present in each group of data to each operational module
Exceptional value, noise figure, redundancy value are rejected, and calculate the mean value of each group of data, by the mean value replace original exceptional value and
Noise figure;The data normalization method is the standard deviation for first calculating each group of data, then each data of each group of data are subtracted
The mean value of this group of data maps the data within [0,1] section then divided by the standard deviation of the group data set.
Further, the step 3 carries out characteristics extraction, extracting method on the basis of step 2 are as follows: extracts each work
Make each group voltage data in module in setting time length, by data phase adduction in operating voltage group in every group of data with
The number of data obtains mean value in group;Operating voltage data subtract mean value and obtain being added after result carries out first square in organizing, so
Variance is obtained divided by data amount check in organizing again afterwards;Variance is opened into radical sign and obtains standard deviation;Position in being found in every group data set again
Number, peak value, valley, each operational module obtain the data set guarantor of a class mean, variance, standard deviation, median, peak value, valley
It is stored in document data set.
Further, the generation of the positive and negative samples data set of the step 4 includes following below scheme:
A. ergodic data collection file individually proposes the data set of putty abnormal conditions, generates the negative sample data of putty exception
Collect file;The normal data set of working condition is proposed, generates the negative sample document data set of normal work, and by two item number
It is saved in document data set according to collection file;
B. in the document data set of putty exception, the random abnormal conditions for taking out setting ratio are used as abnormal subset, and to different
Normal subset is analyzed, and is then added in data set according to the artificial synthesized new sample of abnormal subset sample, the new data set
It is used as sample with train classification models.
1. a kind of injection molding machine putty method for detecting abnormality based on integrated study according to claim 5, feature
Be, in step 5 construct putty anomaly classification device the following steps are included:
A. Logistic classifier is constructed, acquires optimum coefficient vector, including following below scheme using stochastic gradient ascent algorithm:
A. loading of databases file, extraction place characteristic value data collection file and positive and negative samples data set text from document data set
Part;
B. the characteristic value data collection file of extraction and positive and negative samples document data set are upset at random, is divided into according to the ratio of 7:3
Four characteristics: training set feature, test set feature, training set target, test set target;
C. by process b training set feature and training set target switch to matrix-type;
D. Logistic classifier of the component based on stochastic gradient ascent algorithm, setting rise step-length, the number of iterations, initialization system
Number vector then carries out gradient rising under the number of iterations to coefficient vector and acquires optimum coefficient vector;
E. precision test is carried out to training pattern using test set feature and test set target.
B. SVM classifier is constructed, is mapped characteristic value using Radial basis kernel function, using cross validation, selection is most
Good parameter C(penalty factor) and g(kernel function gamma value), with optimal parameter C and g to entire training set training obtain SVM mould
Type, detailed process are as follows:
A. loading of databases file, extraction place characteristic value data collection file and positive and negative samples data set text from document data set
Part;
B. by process a also true data collection and positive and negative samples data set upset at random, be divided into four spies according to the ratio of 7:3
Levy part: training set feature, test set feature, training set target, test set target;
C. Radial basis kernel function (rbf) component SVM classifier is selected, selects the C and g of specific pickup to combine, in training set feature
Make the highest combination of SVM classifier precision with C and g is found in training set target;
D. precision test is carried out to training pattern using test set heat symptom-complex and test set target;
C. decision tree classifier is constructed, according to gini index minimization principle, feature is chosen, building CART decision tree point
Class device, detailed process are as follows:
A. loading of databases file, extraction place characteristic value data collection file and positive and negative samples data set text from document data set
Part;
B. by process a characteristic data set and categorized data set upset at random, be divided into four parts according to the ratio of 7:3: point
It is not training set feature, test set feature, training set target, test set target;
C. decision tree classifier is imported, decision-tree model is generated using training set feature and training set target, in test set feature
With the nicety of grading for verifying model in test set target;
D. CCP(Cost Complexity Pruning is used) method is to the decision-tree model progress beta pruning in step c), from original
Beginning decision treeT 0 Start generate subtree sequenceT 0 , T 1 ... .T n , whereinT i+1 FromT i It generates,T n For root node;
D. increase to the result of the classifier constructed in step A, B, step C as new add in original data set, generate
New data set is passed in the Logistic classifier of the second pole as input, is kept away in the second pole Logistic classifier
Exempt from over-fitting, the training of model carried out using the method for K folding cross validation, detailed process is as follows:
A. training set is divided into K parts;
B. select a copy of it as test set, remaining carries out model training as training set;
C. step b) is repeated, the model for selecting measuring accuracy optimal is as final second level sorter model.
The result of second level Logistic classifier is exported as final result.
Compared with prior art, the invention has the following advantages:
(1) present invention has higher accuracy compared to traditional machine learning model, abnormal be more good putty
Recall rate;
(2) present invention compares traditional detection device, and use is more convenient, and while putty occurs, staff can be timely
It reacts and is repaired by the data set of operating voltage;
(3) present invention is a suitable for accurately and quickly detecting extremely to injection molding machine putty under industrial environment.
Detailed description of the invention
Fig. 1 is the general structure schematic diagram of injection molding machine putty method for detecting abnormality;
Fig. 2 is data set processing schematic;
Fig. 3 is k folding cross validation schematic diagram.
Specific embodiment
In the application system towards injection molding machine putty abnormality detection, using provided by the present invention based on integrated study
Method may be implemented quickly and accurately to find injection molding machine putty extremely, with the injection molding machine putty abnormality detection of certain factory
As shown in Figure 1, specific implementation steps are as follows for system:
1) voltage collector is used, injection molding machine module operating voltage is acquired in real time, acquires A here 0 ~A 1 Three
The voltage of operational module:
Acquisition module sends collected operating voltage to by MQTT protocol channel on the backstage of injection molding machine putty abnormality detection
Program, after background program subscribes to correlation Topic, so that it may real-time reception to acquisition module acquires the operating voltage transmitted,
Spooler will save in the real-time input database of these operating voltages;
2) after acquiring a period of time, collected data are pre-processed, such as Fig. 2, detailed process is as follows:
The first step is that a point calculates separately them with the data in 5 minutes to the voltage data of three operational modules of acquisition
Respective operating voltage mean value in five minutes, with setting operating voltage threshold interval compare, more than threshold interval data just
Regard as abnormal data;
Second step is replaced abnormal data using the average voltage in respective operational module five minutes;
Third step is equally replaced default data using the average voltage in respective operational module five minutes;
3) it extracts characteristic value in the file saved from step 2) to be pre-processed, detailed process is as follows:
The first step traverses the operating voltage of three operational modules, and the operating voltage of each module subtracted and is found out in the first step
Mean value, then quadratic sum is added, and divided by the voltage number of each operational module, is finally opened radical sign and is found out standard deviation;
Second step traverses the operating voltage of three operational modules, and the operating voltage of each module subtracted and is found out in the first step
Mean value, the standard deviation then arranged divided by each operational module voltage, is normalized, maps the data into [0,1] section
Within;
Third step, the data point in traversal five minutes, analyzes the voltage data of three modules, extracts median, peak
Three value, valley characteristic values;
Data after normalized are saved into file by the 4th step again;
For example, in injection molding machine putty abnormality detection, mainly acquisition A 0 ~A 1 The voltage of three operational modules, respectively with 5 points
Zhong Weiyi collection point handles the operating voltage data of acquisition, if acquiring within five minutes 20 data points, three moulds
Just there are 60 data points in block five minutes, this 60 data points are exactly a collection point, utilize mean value formula and variance, standard deviation
Formula finds out mean value, variance and the standard deviation of operating voltage in three operational modules five minutes respectively, then finds out I d median
With peak value, valley, six characteristic values are extracted altogether, and three operational modules realize altogether with regard to 18 characteristic values original 60
The feature of a dimension carries out dimension-reduction treatment;
4) file saved in step 3) is handled, it is ensured that the positive negative sample that initial data is concentrated is more balanced, main side
Method is to take away from minority class using a data subset as an example, then creates similar newly synthesized example.These
The example of synthesis is then added to original data set;Detailed process is as follows:
The first step shares 10000 examples in the file saved in step 3), wherein the example worked normally has 9800, and
The example of putty has 200, extracts the file that 200 examples generate putty exception, is defined as minority class sample set S min ;
Second step traverses the file of putty exception, for each sample x in putty exception, using Euclidean distance as criterion calculation
It arrives minority class sample set S min In all samples distance, obtain its k neighbour;
Third step acquires multiplying power N according to one acquisition ratio of sample imbalance ratio setting to determine, for each exception class
X in sample set randomly chooses several samples from its k neighbour;
4th step, it is assumed that the neighbour selected is X n , new sample is constructed as follows:
For example, this project is exactly that 150 examples are selected from 200 putty examples, then repeatedly aforesaid operations 40 times, altogether
6000 exception examples are generated, in addition original 9800, initial data concentration just has altogether 15800 datas, putty hair
Raw probability also becomes 6000/15800=37.9%, improves the probability of putty appearance;
5) two layers of sorter model are constructed, the classifier of the building first order is first had to, first order classifier includes three, respectively
It is Logistic classifier, SVM classifier and decision tree classifier, second level classifier is a Logistic classifier, tool
Body process is as follows:
The first step constructs Logistic classifier:
(a) data processing module reads text, therefrom reads mean value, variance, standard deviation, median, peak value, six column data of valley
Form the three-dimensional matrice dataMatrix of x;
(b) data processing module reads text, therefrom reads result column data, forms the label matrix labelMatrix of y;
(c) dataMatrix and labelMatrix is generated into training set, training according to randomly selecting in the way of 7:3 respectively
Collect target, four test set, test set target parts;
(d) coefficient column matrix is set as weightMatrix, and length is the number of labelMatrix, weightMatrix with
The result that dataMatrix is multiplied is exactly label matrix labelMatrix;
(e) optimal coefficient matrix column, setting learning rate iteration time are found out using stochastic gradient ascent algorithm on training set
Number loops through until finding optimal coefficient matrix;
(f) trained model is tested on test set.
For example, this project settings learning rate is 0.001, the number of iterations 500.In 500 iteration,
It is the matrix that n train value is 1 that weightMatrix coefficient matrix, which is initialized as value, and the size of n is labelMatrix label matrix
Length randomly selects a certain number of data sets as subset from dataMatrix, constantly by the data subset of extraction and
WeightMatrix is multiplied, and the result after they are multiplied is passed in Sigmoid function as parameter, calculates result square
Battle array, subtracts calculated matrix of consequence for true label matrix labelMatrix, this represent error matrixes, then go more
New coefficient column matrix weightMatrix, calculation method be original coefficient column matrix weightMatrix plus learning rate with
DataMatrix be multiplied with error matrix after matrix.Continuous iteration, until finding optimal coefficient matrix
weightMatrix。
Second step constructs SVM classifier:
(a) by raw data set according to randomly selecting in the way of 7:3 respectively, training set, training set target, test are generated
Four collection, test set target parts;
(b) SVM classifier is constructed using rbf basic function, selectes the section of C and g, constantly chooses different C and g and be combined,
The effect that SVM classifier is verified on training set, selects the combination of optimal C and g;
(c) trained SVM classifier is carried out to the verifying of actual effect on test set and test set target.
This project punishment parameter C setting range is [0.001,0.01,0.1,1,10,20,40,80,100], and g sets model
It encloses for [0.001,0.005,0.1,0.15,0.20,0.25,0.30,0.35], Selection of kernel function is RBF kernel function, is constantly combined
C and g, on the basis of constructing SVM classifier, then the SVM classifier of previous step selection, part is found most in the section for adjusting C again
Advantage, for example if upper step C local best points are when 0.1, the section that C is reset when searching again is
[0.01,0.02,0.04,0.06,0.08,0.1,0.12,0.14,0.16,0.18,0.20] combines the g in previous step again,
Until the C and g that find are optimum combinations.
Third step constructs decision tree classifier:
(a) by raw data set according to randomly selecting in the way of 7:3 respectively, training set, training set target, test are generated
Four collection, test set target parts;
(b) each feature is extracted on training set, forms feature list and the possible classification of each feature is calculated to each feature
Original data set is divided into 2 parts by situation, then calculates Gini index;
(c) according to the smallest feature of Gini and corresponding cut-off as optimal characteristics and optimal cut-off, according to optimal
Feature and optimal cut-off generate child node from existing node, training dataset are assigned to two child nodes according to feature
In;
(d) to two child node recursive calls, two child nodes, training dataset is assigned in two child nodes by feature
It goes;
(e) last CART decision-tree model is generated, is verified on test set.
For example, this items selection is CART decision tree, therefore use Gini value as judgment criteria.Formula is:, wherein p (i) is the ratio of i-th of sample in data set on present node.This
Project is two classification, it is assumed that classification samples have 1000, and wherein the sample of the first kind has 700, and the sample of the second class has 300
It is a, then Gini=1-0.7*0.7-0.3*0.3=0.42, it can be seen that category distribution is more uniform, and Gini value is bigger, therefore, differentiates
Standard is exactly to find optimal feature and threshold value, so that the Gini value of present node subtracts the result of the Gini value of left and right node most
Greatly.
6) it is passed to second level classifier using three output results of first order classifier as input, detailed process is such as
Under:
The first step, select in the result of three classifiers of the first order the highest result of probability of occurrence for first order classifier most
Eventually as a result, being added in original data set using the final result of first order classifier as newly adding;
Second step, in order to avoid over-fitting, the method such as Fig. 3 of the Logistic classifier of the second level using n folding cross validation, number
Cross validation training pattern is rolled over by n according to collection, the specific steps are as follows:
(a) original data set is divided into n parts (value of general n is 2-10);
(b) in d 1 ,d 2 …..d n Middle random selection portion d i As test set, remaining is trained as training set, records
Accuracy;
(c) step (b) is repeated, selects that highest model of accuracy as final model.
Third step, after the completion of training pattern, the data set that the first step generates later can directly be passed to the second level and classify
Last division is carried out in device.
Finally, it should be noted that the above embodiments are merely illustrative of the technical solutions of the present invention, rather than its limitations;Although
Present invention has been described in detail with reference to the aforementioned embodiments, it will be understood by those of skill in the art that it still can be right
Technical solution documented by foregoing embodiments is modified, or is replaced on an equal basis to part of technical characteristic;And this
It modifies or replaces, the spirit and model of technical solution of various embodiments of the present invention that it does not separate the essence of the corresponding technical solution
It encloses.
Claims (6)
1. a kind of injection molding machine putty method for detecting abnormality based on integrated study, which is characterized in that method includes the following steps:
1) raw data acquisition
Module voltage collector is installed on each operational module of injection molding machine, and each work of timing acquiring within the set duration
The operating voltage made in module is stored as initial data into database;
2) data prediction
Data are successively used to operating voltage each group of data of each operational module stored in database within the assigned work time
Cleaning method, data normalization method pretreatment;
3) characteristics extraction
Pretreated data are obtained to step 2 and extract characteristic value, composition with temporal signatures (mean value, variance, standard deviation, in
Digit, peak value, valley) data set, and store into document data set;
4) data set positive and negative samples are generated
New data set in data file is divided into putty abnormal data set and works normally data set, and using synthesis minority class
Oversampling technique (SMOTE) increases the data set of putty abnormal conditions, the positive and negative samples that equilibrium data is concentrated;
5) putty anomaly classification device is constructed
Secondary classifier is constructed on the basis of step 4, the first order is three base classifiers, and the second level is a classifier, the
Three base classifiers of level-one are respectively Logistic classifier, SVM classifier and decision tree classifier, and second level classifier is
Logistic classifier.
2. a kind of injection molding machine putty method for detecting abnormality based on integrated study according to claim 1, which is characterized in that
When periodically acquiring the voltage data of each operational module of injection molding machine in setting duration section in the step 1, need in real time
The product specification of injection molding machine production is recorded, once there is the period of putty exception labeled as 1, remaining, which is worked normally, is labeled as 0.
3. a kind of injection molding machine putty method for detecting abnormality based on integrated study according to claim 2, which is characterized in that
Data cleaning method in the step 2 is exceptional value, noise figure, redundancy present in each group of data to each operational module
Value is rejected, and calculates the mean value of each group of data, which is replaced to original exceptional value and noise figure;The data normalizing
Change method is first to calculate the standard deviation of each group of data, then each data of each group of data are subtracted to the mean value of this group of data, then
Divided by the standard deviation of the group data set, map the data within [0,1] section.
4. a kind of injection molding machine putty method for detecting abnormality based on integrated study according to claim 2 or 3, feature exist
In the step 3 carries out characteristics extraction, extracting method on the basis of step 2 are as follows: when extracting setting in each operational module
Between each group voltage data in length, by the number for organizing interior data at data phase adduction in operating voltage group in every group of data
Obtain mean value;Operating voltage data subtract mean value and obtain being added after result carries out first square in organizing, then again divided by number in group
Variance is obtained according to number;Variance is opened into radical sign and obtains standard deviation;Median, peak value, valley are found in every group data set again, often
A operational module obtain a class mean, variance, standard deviation, median, peak value, valley data set be saved in document data set
In.
5. a kind of injection molding machine putty method for detecting abnormality based on integrated study according to claim 4, which is characterized in that
The generation of the positive and negative samples data set of the step 4 includes following below scheme:
A. ergodic data collection file individually proposes the data set of putty abnormal conditions, generates the negative sample data of putty exception
Collect file;The normal data set of working condition is proposed, generates the negative sample document data set of normal work, and by two item number
It is saved in document data set according to collection file;
B. in the document data set of putty exception, the random abnormal conditions for taking out setting ratio are used as abnormal subset, and to different
Normal subset is analyzed, and is then added in data set according to the artificial synthesized new sample of abnormal subset sample, this is new
Data set is used as sample with train classification models.
6. a kind of injection molding machine putty method for detecting abnormality based on integrated study according to claim 5, which is characterized in that
In step 5 construct putty anomaly classification device the following steps are included:
A. Logistic classifier is constructed, acquires optimum coefficient vector, including following below scheme using stochastic gradient ascent algorithm:
A. loading of databases file, extraction place characteristic value data collection file and positive and negative samples data set text from document data set
Part;
B. the characteristic value data collection file of extraction and positive and negative samples document data set are upset at random, is divided into according to the ratio of 7:3
Four characteristics: training set feature, test set feature, training set target, test set target;
C. by process b training set feature and training set target switch to matrix-type;
D. Logistic classifier of the component based on stochastic gradient ascent algorithm, setting rise step-length, the number of iterations, initialization system
Number vector then carries out gradient rising under the number of iterations to coefficient vector and acquires optimum coefficient vector;
E. precision test is carried out to training pattern using test set feature and test set target.
B. SVM classifier is constructed, is mapped characteristic value using Radial basis kernel function, using cross validation, selects best ginseng
Number C(penalty factors) and g(kernel function gamma value), with optimal parameter C and g to entire training set train obtain SVM model,
Detailed process is as follows:
A. loading of databases file, extraction place characteristic value data collection file and positive and negative samples data set text from document data set
Part;
B. by process a also true data collection and positive and negative samples data set upset at random, be divided into four spies according to the ratio of 7:3
Levy part: training set feature, test set feature, training set target, test set target;
C. Radial basis kernel function (rbf) component SVM classifier is selected, selects the C and g of specific pickup to combine, in training set feature
Make the highest combination of SVM classifier precision with C and g is found in training set target;
D. precision test is carried out to training pattern using test set heat symptom-complex and test set target;
C. decision tree classifier is constructed, according to gini index minimization principle, feature is chosen, building CART decision tree point
Class device, detailed process are as follows:
A. loading of databases file, extraction place characteristic value data collection file and positive and negative samples data set text from document data set
Part;
B. by process a characteristic data set and categorized data set upset at random, be divided into four parts according to the ratio of 7:3: point
It is not training set feature, test set feature, training set target, test set target;
C. decision tree classifier is imported, decision-tree model is generated using training set feature and training set target, in test set feature
With the nicety of grading for verifying model in test set target;
D. CCP(Cost Complexity Pruning is used) method is to the decision-tree model progress beta pruning in step c), from original
Beginning decision treeT 0 Start generate subtree sequenceT 0 , T 1 ... .T n , whereinT i+1 FromT i It generates,T n For root node;
D. increase to the result of the classifier constructed in step A, B, step C as new add in original data set, generate
New data set is passed in the Logistic classifier of the second pole as input, is kept away in the second pole Logistic classifier
Exempt from over-fitting, the training of model carried out using the method for K folding cross validation, detailed process is as follows:
A. training set is divided into K parts;
B. select a copy of it as test set, remaining carries out model training as training set;
C. step b) is repeated, the model for selecting measuring accuracy optimal is as final second level sorter model.
The result of second level Logistic classifier is exported as final result.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810790797.8A CN109145948A (en) | 2018-07-18 | 2018-07-18 | A kind of injection molding machine putty method for detecting abnormality based on integrated study |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810790797.8A CN109145948A (en) | 2018-07-18 | 2018-07-18 | A kind of injection molding machine putty method for detecting abnormality based on integrated study |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109145948A true CN109145948A (en) | 2019-01-04 |
Family
ID=64801136
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810790797.8A Withdrawn CN109145948A (en) | 2018-07-18 | 2018-07-18 | A kind of injection molding machine putty method for detecting abnormality based on integrated study |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109145948A (en) |
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109799379A (en) * | 2019-01-11 | 2019-05-24 | 厦门南鹏物联科技有限公司 | Method for measuring charged, charging detection device and socket |
CN110163381A (en) * | 2019-04-26 | 2019-08-23 | 美林数据技术股份有限公司 | Intelligence learning method and device |
CN110163442A (en) * | 2019-05-27 | 2019-08-23 | 华北理工大学 | A kind of gas well plug-ging prediction technique based on integrated study |
CN110309955A (en) * | 2019-06-13 | 2019-10-08 | 南瑞集团有限公司 | A kind of non-load predicting method and device shut down when upgrading of cloud environment application system |
CN110609530A (en) * | 2019-09-23 | 2019-12-24 | 厦门华夏国际电力发展有限公司 | Data mining method and system for realizing working condition optimization based on DCS system edge |
CN111312329A (en) * | 2020-02-25 | 2020-06-19 | 成都信息工程大学 | Transcription factor binding site prediction method based on deep convolution automatic encoder |
CN112836645A (en) * | 2021-02-04 | 2021-05-25 | 浙江工业大学 | Large-scale exercise heart rate sequence-oriented running-instead detection method |
CN113515678A (en) * | 2021-05-13 | 2021-10-19 | 上海梯之星信息科技有限公司 | Abnormal data screening method |
CN113807418A (en) * | 2021-09-02 | 2021-12-17 | 乐创达投资(广东)有限公司 | Injection molding machine energy consumption abnormity detection method and system based on Gaussian mixture model |
WO2023208136A1 (en) * | 2022-04-28 | 2023-11-02 | 郑州云海信息技术有限公司 | Kpi anomaly detection method and apparatus, device and medium |
-
2018
- 2018-07-18 CN CN201810790797.8A patent/CN109145948A/en not_active Withdrawn
Cited By (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109799379B (en) * | 2019-01-11 | 2022-01-11 | 厦门南鹏物联科技有限公司 | Charging detection method, charging detection device and socket |
CN109799379A (en) * | 2019-01-11 | 2019-05-24 | 厦门南鹏物联科技有限公司 | Method for measuring charged, charging detection device and socket |
CN110163381A (en) * | 2019-04-26 | 2019-08-23 | 美林数据技术股份有限公司 | Intelligence learning method and device |
CN110163442A (en) * | 2019-05-27 | 2019-08-23 | 华北理工大学 | A kind of gas well plug-ging prediction technique based on integrated study |
CN110309955A (en) * | 2019-06-13 | 2019-10-08 | 南瑞集团有限公司 | A kind of non-load predicting method and device shut down when upgrading of cloud environment application system |
CN110309955B (en) * | 2019-06-13 | 2022-07-15 | 南瑞集团有限公司 | Load prediction method and device during non-shutdown upgrading of cloud environment application system |
CN110609530A (en) * | 2019-09-23 | 2019-12-24 | 厦门华夏国际电力发展有限公司 | Data mining method and system for realizing working condition optimization based on DCS system edge |
CN111312329A (en) * | 2020-02-25 | 2020-06-19 | 成都信息工程大学 | Transcription factor binding site prediction method based on deep convolution automatic encoder |
CN111312329B (en) * | 2020-02-25 | 2023-03-24 | 成都信息工程大学 | Transcription factor binding site prediction method based on deep convolution automatic encoder |
CN112836645A (en) * | 2021-02-04 | 2021-05-25 | 浙江工业大学 | Large-scale exercise heart rate sequence-oriented running-instead detection method |
CN112836645B (en) * | 2021-02-04 | 2024-03-29 | 浙江工业大学 | Substitution running detection method for large-scale exercise heart rate sequence |
CN113515678A (en) * | 2021-05-13 | 2021-10-19 | 上海梯之星信息科技有限公司 | Abnormal data screening method |
CN113807418A (en) * | 2021-09-02 | 2021-12-17 | 乐创达投资(广东)有限公司 | Injection molding machine energy consumption abnormity detection method and system based on Gaussian mixture model |
WO2023208136A1 (en) * | 2022-04-28 | 2023-11-02 | 郑州云海信息技术有限公司 | Kpi anomaly detection method and apparatus, device and medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109145948A (en) | A kind of injection molding machine putty method for detecting abnormality based on integrated study | |
Ru et al. | Interpretable neural architecture search via bayesian optimisation with weisfeiler-lehman kernels | |
CN109873610B (en) | Photovoltaic array fault diagnosis method based on IV characteristic and depth residual error network | |
CN110597240B (en) | Hydroelectric generating set fault diagnosis method based on deep learning | |
CN106845717B (en) | Energy efficiency evaluation method based on multi-model fusion strategy | |
CN108875771B (en) | Fault classification model and method based on sparse Gaussian Bernoulli limited Boltzmann machine and recurrent neural network | |
CN111444940A (en) | Fault diagnosis method for critical parts of fan | |
CN108875772B (en) | Fault classification model and method based on stacked sparse Gaussian Bernoulli limited Boltzmann machine and reinforcement learning | |
US11860608B2 (en) | Industrial equipment operation, maintenance and optimization method and system based on complex network model | |
CN108051660A (en) | A kind of transformer fault combined diagnosis method for establishing model and diagnostic method | |
US20210334658A1 (en) | Method for performing clustering on power system operation modes based on sparse autoencoder | |
CN113688869B (en) | Photovoltaic data missing reconstruction method based on generation countermeasure network | |
CN110717610A (en) | Wind power prediction method based on data mining | |
CN108717149A (en) | Diagnosis Method of Transformer Faults based on M-RVM fusion dynamic weightings AdaBoost | |
CN108647707B (en) | Probabilistic neural network creation method, failure diagnosis method and apparatus, and storage medium | |
CN113469230B (en) | Rotor system deep migration fault diagnosis method, system and medium | |
CN116842337A (en) | Transformer fault diagnosis method based on LightGBM (gallium nitride based) optimal characteristics and COA-CNN (chip on board) model | |
CN117458480A (en) | Photovoltaic power generation power short-term prediction method and system based on improved LOF | |
CN113705887A (en) | Data-driven photovoltaic power generation power prediction method and system | |
CN116245259B (en) | Photovoltaic power generation prediction method and device based on depth feature selection and electronic equipment | |
CN117151488A (en) | Method, system, storage medium and equipment for expanding cold tide and strong wind weather sample | |
CN116720095A (en) | Electrical characteristic signal clustering method for optimizing fuzzy C-means based on genetic algorithm | |
Bo | Research on the classification of high dimensional imbalanced data based on the optimizational random forest algorithm | |
CN113496255B (en) | Power distribution network mixed observation point distribution method based on deep learning and decision tree driving | |
US20220138631A1 (en) | Systems and methods for photovoltaic fault detection using a feedback-enhanced positive unlabeled learning |
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 | ||
WW01 | Invention patent application withdrawn after publication |
Application publication date: 20190104 |
|
WW01 | Invention patent application withdrawn after publication |