CN109902954A - A kind of flexible job shop dynamic dispatching method based on industrial big data - Google Patents
A kind of flexible job shop dynamic dispatching method based on industrial big data Download PDFInfo
- Publication number
- CN109902954A CN109902954A CN201910144370.5A CN201910144370A CN109902954A CN 109902954 A CN109902954 A CN 109902954A CN 201910144370 A CN201910144370 A CN 201910144370A CN 109902954 A CN109902954 A CN 109902954A
- Authority
- CN
- China
- Prior art keywords
- scheduling
- data
- rule
- machine
- workpiece
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
A kind of flexible job shop dynamic dispatching method based on industrial big data, includes the following steps as follows: step 1: Usage data collection tool Sqoop and Flume, scheduling data is obtained from database or file system, and be stored in HDFS file system；Step 2: it by Tool for Data Warehouse Hive, is divided for dispatching data by unit of scheduling scheme；Step 3: it is converted into trained example by data are dispatched using Spark Computational frame, and is stored in the form of scheduling scheme is unit into Hbase；Step 4: being screened from multiple indexs, obtains showing the scheduling data acquisition system that good scheduling scheme generates in commission；Step 5: the cluster based on disturbance attribute is carried out to scheduling relevant historical data；Step 6: using improved random forests algorithm, excavates random forest scheduling rule；Step 7: flexible job shop dynamic dispatching is instructed using excavated scheduling rule.Practical operability of the present invention, computational efficiency are high, quickly can make real-time response to workshop disturbance.
Description
Technical field
The present invention relates to a kind of flexible job shop dynamic dispatching methods based on industrial big data
Background technique
Scheduling plays an important role in manufacture system, and scheduling quality will will affect the competitiveness of manufacturing enterprise itself.
Formulating scientific and reasonable scheduling scheme for workshop can be improved production efficiency, reduction process cost, shortening product life cycle, together
When can guarantee to deliver goods with guaranteeing the quality on time.Flexible job shop due to its with flexible process route with to the fast of the market demand
Fast adaptability to changes can be very good the production requirement for meeting multi items, small lot, and therefore, it has become a kind of lifes being widely used
Production mode.Flexible job shop dynamic dispatching is that the disturbance of actual production environment is considered on the basis of static scheduling, is more accorded with
Actual production environment is closed, so more having research significance.
As product demand constantly changes to personalization, manufacturing process is more various, and actual schedule problem also becomes more
Complexity, manufacturing enterprise is to Job-Shop way to solve the problem in practical operability, computational efficiency and to workshop disturbance
Real-time response ability etc. made higher requirement.Priority scheduling rule is a kind of simple heuristic rule, it
Computational efficiency is high, practical strong operability and can be used for Real-Time Scheduling, is suitable for the dynamic dispatch environment of complexity.But priority scheduling
The performance of rule is influenced by actual environment variation, and single scheduling rule cannot have preferably in all bumpy weathers
Scheduling performance.In order to meet the needs of actual job Job-Shop, a kind of feasible thinking is from scheduling relevant historical data
It is middle to excavate the scheduling knowledge about scheduling rule to instruct practical Job-Shop activity.Scheduling problem is solved by data mining
Research be broadly divided into the method in conjunction with existing priority scheduling rule and the excavation scheduling rule from scheduling relevant historical data
Method.
In terms of combining existing priority scheduling rule, WANG Shuang-Xi etc. is in (A hybridknowledge
discovery model using decision tree and neural network for selecting
Dispatching rules of a semiconductor final testing factory, 2005) it is directed to semiconductor row
Industry proposes a kind of combination decision tree and neural network and excavates priority scheduling rule selection mechanism from scheduling relevant historical data
Method, most suitable priority scheduling rule under the available current environment of the selection mechanism.SHIUE Y.R. etc. is in (Data-
mining-based dynamic dispatching rule selection mechanism for shop floor
Control systems using a support vector machine approach, 2009) it proposes a kind of using supporting
Vector machine (SupportVector Machine, SVM) excavates priority scheduling rule selection mechanism from scheduling relevant historical data
Method, and real-time operation decision is made with this.Mouelhi is in (Training a neuralnetwork to select
Dispatching rules in real time, 2009) etc. a kind of scheduling rule selecting party of combination neural network is proposed
Method, this method excavate scheduling rule real-time selection method by neural network from the scheduling relevant historical data that emulation generates.
Priority scheduling rule only makes scheduling decision by a small amount of information, this may cause scheduling result not to the utmost such as people
Meaning, therefore extracting new scheduling rule from scheduling relevant historical data is another thinking.LI X etc. is in (Discovering
Dispatching rules using datamining, 2005) propose it is a kind of using decision tree from scheduling relevant historical data
In obtain the method for completely new scheduling rule, and be experimentally confirmed extract scheduling rule can be fitted former dispatching party well
Case.SIGURDUR OLAFSSON etc. is in (Learning effective new single machine dispatching
Rules from optimal scheduling data, 2010) a kind of two stage scheduling knowledge learning method is proposed, it is first
First learn the training example for how obtaining being suitble to excavation from scheduling process data, then rule is scheduled to these training examples
Excavation.Wang Chenglong etc. proposes a kind of combination Petri network (method for digging of solving job shop scheduling problem rule is studied, 2015) and builds
The branch-bound algorithm of mould and the scheduling rule method for digging of decision Tree algorithms, the scheduling rule of extraction can be used for instructing static work
Industry Job-Shop.
In conclusion being primarily directed to vehicle about the method for excavating scheduling rule from scheduling relevant historical data at present
Between on scheduling problem, applied to the less of flexible job shop dynamic scheduling problem.In addition, the above method is used
Scheduling relevant historical data be partial to gross data, a large amount of uses however as Intellisense equipment in shop layer, workshop
Start to intelligent development, Job-Shop relevant historical data shows the industry such as scale is big, value is low, continuous sampling, higher-dimension
The characteristics of big data.
Summary of the invention
In order to overcome, the existing practical operability of flexible job shop dynamic dispatching method is not high, computational efficiency is completely and to vehicle
Between disturb real-time response scarce capacity the problem of, the present invention provide a kind of practical strong operability, computational efficiency it is high, can be right
Workshop disturbs the flexible job shop dynamic dispatching method for making real-time response.
The technical solution adopted by the present invention to solve the technical problems is:
A kind of flexible job shop dynamic dispatching method based on industrial big data, the flexible job shop dynamic dispatching
Method the following steps are included:
Step 1, data acquisition: using the metadata acquisition tool under Hadoop ecosystem from existing information system
Historical data relevant to scheduling is acquired, and is stored in HDFS file system.
Step 2, Data Integration: by Tool for Data Warehouse Hive using SQL statement to the scheduling in HDFS file system
Data acquisition system DhIt is divided by unit of scheduling scheme, i.e., the scheduling relevant historical generated primary scheduling scheme in commission
Data are divided into together.
Step 3, the data relay of integration: being turned to the form of trained example using Spark by data conversion, is convenient for data
Mining algorithm is scheduled rule digging.
Step 4, data screening: from three Maximal Makespan, total tardiness time, machine total load indexs to history tune
Degree scheme is considered, and screening obtains showing the scheduling relevant historical data set that good scheduling scheme generates in commission.
It specifically includes:
Step 4.1: in Maximal Makespan index, to only use the dispatching party of SPT rule generation in situation of the same race
The Maximal Makespan of case is as screening criteria.
Step 4.2: identical to use EDD rule combination SPT rule to complete in situation of the same race in total tardiness time index
In the case of scheduler task total extension time as screening criteria.
Step 4.3: in machine total load index, to combine LMWT and SPT rule to complete scheduler task in situation of the same race
Machine total load as screening criteria, the scheduling data that the scheduling scheme of these three indexs generates in commission can be met simultaneously
Set, it will the input as scheduling rule mining algorithm.
Step 5；Cluster based on disturbance attribute: DBSCAN clustering method is used, to the scheduling relevant historical number after screening
According to carrying out based on disturbance attribute using scheduling scheme as unit (i.e. the data that a scheduling scheme generates are as an object)
Cluster.It specifically includes:
Step 5.1: noisy data when executing to scheme carries out data normalization processing, if certain disturbance attribute is in each side
Data in case are X1, X2, X3..., Xn, then they need to convert by such as formula (1).
In formula (1)Indicate the mean value of attribute；S is expressed as standard deviation；Y1, Y2, Y3..., Yn are the data after standardization.
Step 5.2: determining the parameter field radius Eps of DBSCAN algorithm, and included at least in the radius of kernel object field
Object number MinPts.
Step 5.3: finding out a kernel object p at random, create new cluster of the p as kernel object.It is found again from p
The reachable object of direct density, is classified in cluster.
Step 5.4: repeating step 5.3, when not new point can be added to any cluster, which terminates.
Step 6, random forest scheduling rule are excavated: being used improved random forests algorithm, divided from cluster each after cluster
The forest scheduling scheduling rule 1 that workpiece selection machine problem Wa Jue be resolved is asked with idle machine selection workpieces processing is solved
The random forest scheduling 2 of topic.It specifically includes:
Step 6.1: for the cluster after each cluster, extracting trained example with putting back to from cluster, form k new instructions
Practice example collection, for constructing decision tree.
Step 6.2: m characteristic attribute of random selection, and best divisional mode is calculated, it is respectively trained to obtain k decision tree.
Step 6.3: using non-selected trained example in cluster, the classification performance of test decision tree.
Step 6.4: judging whether there is similar decision tree, if it exists similar decision tree, then reservation table is existing in testing
Good decision tree, forms random forest.
Step 6.5: logical to calculate the weight w of every decision tree according to Bayes's voting mechanism, h obtains forest scheduling scheduling
Rule 1 and random forest scheduling 2.
Step 7, scheduling rule use: instructing flexible job shop dynamic to adjust by excavated random forest scheduling rule
Degree.It specifically includes:
Step 7.1 selects workpieces processing problem according to the problem of solution for workpiece machine choice or idle machine,
It finds and random forest scheduling rule 1 corresponding to cluster belonging to the bumpy weather of current flexible job shop or random forest tune
Metric then 2.
Step 7.2, according to selected random forest scheduling rule, optimal method is selected by comparing two-by-two, in candidate machine
Most suitable workpiece or machine are selected in device set M or candidate artifacts set J.
Technical concept of the invention are as follows: Job-Shop relevant historical data show scale is big, value is low, continuous sampling,
The characteristics of industry big data such as higher-dimension, therefore combine big data according to the pretreatment for completing scheduling relevant historical data first.Fig. 2 is provided
Combine the data prediction model of big data technology.Data prediction flexible job shop dynamic scheduling problem is that having disturbance
In the environment of the data acquisition system D that solves the problems, such as the machine choice of workpiece and the workpiece select permeability of idle machine, therefore acquirehPoint
It is system disturbance information relevant to disturbance when scheduling scheme is formulated for three parts: d1；D2 is that certain procedure selection of workpiece adds
When work machine, the status information of every machine in the collection of machines of this procedure can be currently processed；D3 needs for idle machine
When selecting workpiece to be processed in waiting list, the status information of each workpiece in current queue.Dispatch data acquisition system
DhIn data mode it is chaotic, cannot be used directly for next data screening, cluster and scheduling rule excacation, need
Pass through Data Integration and the arrangement scheduling data acquisition system D that is convertedhIn data.In scheduling data acquisition system DhIn to imply reflection real
The mass efficient information of border dispatch environment feature and scheduling knowledge, at the same also along with many useless or wrong rule or
Mode.Therefore the multiple parameter data Filtering system of Fig. 3 is used, from three Maximal Makespan, total tardiness time, machine total load angles
Degree considers history scheduling scheme, retains generated data in the schedule history scheme execution for meet three indexs.
Using random forests algorithm as the mining algorithm of scheduling rule, the algorithm building thus of finally obtained scheduling rule
Random forest, is essentially more trained C4.5 decision trees, and the scheduling performance of scheduling rule depends on point of decision tree
Class performance, the computational efficiency and complexity of scheduling rule depend on the numbers of branches of decision tree.By DBSCAN cluster to compared with
Excellent scheduling data DbIt is reasonably divided, distinguishes data caused by the scheduling decision made under different bumpy weathers,
The scheduling rule for different bumpy weathers is respectively obtained from each region of division again, gained random forest tune can be enhanced
Metric then in decision tree classification performance and reduce numbers of branches so that scheduling rule complexity is lower, computational efficiency
Higher, scheduling performance is more preferable.
Learning scheduling rule f, f from schedule history related data by random forests algorithm is to true scheduling rule in fact
A kind of then estimation of ySoIt is that there is a certain error between y.Error includes three parts: noise δ2, varianceAnd deviationWherein noise δ2It is inevitable, but can be by reducing varianceOr deviationThe error of algorithm is reduced, to improve the performance of random forests algorithm.The correlation between decision tree is reduced simultaneously
ρ can reduce variance, therefore if the similarity between two decision numbers is excessive, retain the decision number that test does very well, thus
Reduce the correlation ρ between decision tree.Traditional random forests algorithm is using the voting mechanism that the minority is subordinate to the majority, random gloomy
No matter fine or not the classification performance of decision tree is in woods, weight all having the same.Such mechanism results in determining for classification performance difference
Plan tree and the good decision tree of classification performance influence degree having the same for final result.Therefore Bayes's voting machine is used herein
System.The mechanism is classified in testing based on every decision tree shows one weight of setting, then votes according to this weight.
Beneficial effects of the present invention are mainly manifested in: being dug from the scheduling relevant historical data with industrial big data
Scheduling rule is dug to instruct the method for scheduling as main body frame, the data prediction model for combining big data technology is established, mentions
The high speed and accuracy of data prediction, establishes the behavior aggregate based on disturbance attribute, reduces scheduling rule complexity
Degree, improve scheduling rule computational efficiency is higher and scheduling performance, establish based on the scheduling for improving random forests algorithm
Mining model improves the generalization ability and scheduling performance of scheduling rule.
Detailed description of the invention
Fig. 1 is that scheduling rule of the invention excavates overall architecture.
Fig. 2 is the scheduling data prediction model of combination big data technology of the invention.
Fig. 3 is multiple parameter data Filtering system of the invention.
Fig. 4 is the flow chart that improved random forests algorithm of the invention excavates scheduling rule.
Fig. 5 is the resulting dispatching party of flexible job shop dynamic dispatching method of use of the invention based on industrial big data
Case.
Specific embodiment
- Fig. 5 referring to Fig.1, a kind of flexible job shop dynamic dispatching method based on industrial big data, overall framework reference
Fig. 1 is specifically divided into three parts: first part, specific referring to Fig. 2 in conjunction with the scheduling data prediction model of big data technology
It is divided into data acquisition, Data Integration, data conversion and data screening；First part, based on disturbance hierarchical cluster attribute strategy；Third portion
Point, based on the scheduling rule mining model for improving random forests algorithm.Its technical step generally is as follows: step 1, data
Acquisition: existing from MES, ERP, SCADA etc. using the metadata acquisition tool Sqoop and Flume under Hadoop ecosystem
Historical data relevant to scheduling is acquired in information system, and is stored in HDFS file system.Acquiring data includes three
Divide Dh={ d1, d2, d3 }: d1 is system disturbance information relevant to disturbance when scheduling scheme is formulated；D2 is certain road work of workpiece
When sequence selects processing machine, the status information of every machine in the collection of machines of this procedure can be currently processed；D3 is sky
When not busy machine needs to select workpiece to be processed in waiting list, the status information of each workpiece in current queue.
Step 2, Data Integration: by Tool for Data Warehouse Hive using SQL statement to the scheduling in HDFS file system
Data acquisition system DhIt is divided by unit of scheduling scheme, i.e., the scheduling relevant historical generated primary scheduling scheme in commission
Data are divided into together.
Data conversion: d2 and d3 in data after integration is partially converted into the shape of trained example using Spark by step 3
Formula is scheduled rule digging convenient for data mining algorithm.It specifically includes:
Step 3.1: for the scheduling data acquisition system D of acquisitionhThe part d2, by actual selection in certain history scheduling scheme
Machine m1 is considered as most suitable machine, and the machine in the alternative collection of machines { m2, m3... } of this process is can be processed with other in it
Compare to form trained example one by one.
Step 3.2: for the scheduling data acquisition system D of acquisitionhThe part d3, by actual selection in certain history scheduling scheme
Workpiece j1 is considered as most suitable machine, by its with other etc. workpiece in workpiece set { j2, j3... } to be processed compare one by one
Form training example.
Step 4, data screening: from three Maximal Makespan, total tardiness time, machine total load indexs to history tune
Degree scheme is considered, and screening obtains showing the scheduling relevant historical data set that good scheduling scheme generates in commission
Db.It specifically includes:
Step 4.1: in Maximal Makespan index, to only use the dispatching party of SPT rule generation in situation of the same race
The Maximal Makespan of case is as screening criteria.It only uses SPT rule and refers to that most fast machine and idle machine are processed in workpiece selection
Device selects process time shortest workpiece.The scheduling scheme that Maximal Makespan meets this index enters step 4.2, is unsatisfactory for then
It eliminates.
Step 4.2: identical to use EDD rule combination SPT rule to complete in situation of the same race in total tardiness time index
In the case of scheduler task total extension time as screening criteria.SPT+EDD rule refers to that workpiece selection processing is most fast
Machine and idle machine selection delivery date earliest workpiece.The scheduling scheme that total tardiness time meets this index enters step
4.3, it is unsatisfactory for, eliminates.
Step 4.3: in machine total load index, to combine LMWT and SPT rule to complete scheduler task in situation of the same race
Machine total load as screening criteria, LMWT+SPT rule refers to workpiece selection free time longest machine and idle
Machine choice process time shortest workpiece.The scheduling number that the scheduling scheme of these three indexs generates in commission can be met simultaneously
According to set, it will the input as scheduling rule mining algorithm.
Step 5；Cluster based on disturbance attribute: using DBSCAN to DbUsing scheduling scheme as a unit (i.e. dispatching party
The data that case generates are as an object), according to DbIn solution formulation when system disturbance attribute (part d1) be based on
Disturb the cluster of attribute.It specifically includes:
Step 5.1: to d1 partial data standardization, if certain data of disturbance attribute in each scheme is X1, X2,
X3..., Xn, then they need to convert by such as formula (1).
In formula (1)Indicate the mean value of attribute；S is expressed as standard deviation；Y1, Y2, Y3..., Yn are the data after standardization.
Step 5.2: determining the parameter field radius Eps of DBSCAN algorithm, and included at least in the radius of kernel object field
Object number MinPts.
Step 5.3: finding out the kernel object p of one not processed (not being classified as some cluster or labeled as noise) at random
(the object number for including in the radius of field is not less than MinPts), new cluster C is established, all objects in p radius of neighbourhood Eps are added
Enter Candidate Set N.
Step 5.4: finding out not yet processed object q in a Candidate Set N at random.If q is kernel object, by q neighbour
Object that is not processed and not being added to N is added in N in the radius Eps of domain.If q is not included into any one cluster, q is added
Enter C.
Step 5.5: step 5.4 is repeated, until N is sky.
Step 5.6: repeating step 5.3,5.4,5.5, when not new object can be added to any cluster, the mistake
Journey terminates
Step 6, random forest scheduling rule are excavated: being used improved random forests algorithm, divided from cluster each after cluster
The forest scheduling rule 1 that workpiece selection machine problem Wa Jue be resolved selects workpieces processing with idle machine is solved the problems, such as
Random forest scheduling 2.It specifically includes:
Step 6.1: for the cluster after each cluster, (being excavated from the d2 (excavating random forest scheduling rule 1) in cluster with d3
Random forest scheduling rule 2) in extract trained example with putting back to, be respectively formed new training the example collection P1 and P2 of k, be used for
Construct decision tree.
Step 6.2:P1 and P2 randomly chooses m characteristic attribute from d2 and d3 respectively, and calculates best divisional mode, point
K decision tree T1 and T2 Xun Lian not obtained.
Wherein decision tree building process are as follows:
Step 6.2.1: creation root node N.
Step 6.2.2: whether there are also remaining training examples for training of judgement example collection, if the return node N without if, if having
Then in next step.
Step 6.2.3: whether the scheduling decision of training of judgement example collection residue training example is all C, if then returning
Node N, and it is labeled as class C, if having in next step.
Step 6.2.4: judge to produce attribute list whether be it is empty, it is empty then be labeled as most classes occur in sample, otherwise
In next step.
Step 6.2.5: check whether the attribute in Attribute class table is continuity, and connection attribute will be obtained by dichotomy
The maximum attribute separate mode of attribute gain G (D, A).(all properties value of attribute can be divided by two parts by dichotomy,
This shared N-1 kind division methods, the division threshold value of dichotomy are selection two respectively in the average value of adjacent two o'clock.Information gain meter
Calculation mode such as formula (2), (3), (4)).
G (D, A)=H (D)-H (D | A) (2)
G (D, A) indicates the information gain of attribute A in formula (2)；H (D, A) classification information entropy in formula (3)；H in formula (4) (D |
A conditional entropy) is indicated；In addition, D indicates training Exemplar Data Set, | D | indicate the training example quantity of D, and D has K classification
Ck, k=1,2；|Ck| it indicates in classification CkIn training example number.D can be divided into n subset D by attribute A1,
D2..., Dn, | Di| it is DiTraining example number.DiIn belong to class CkThe collection of training example be combined into Dik, | Dik| it is DikInstruction
Practice example number.
Step 6.2.6: the selection maximum attribute flag node N of information gain-ratio, information gain-ratio calculation formula such as formula
(5), (6), return step 6.2.2.
GR (D, A)=G (D, A)/H (A) (5)
GR (D, A) indicates information gain-ratio in formula (5)；H (A) indicates division information；Other symbol meanings are same as above.
Step 6.3: using non-selected trained example in d2 and d3, testing the classification chart of decision tree in T1 and T2 respectively
It is existing.
Step 6.4: calculate the similarity S in T1 or T2 between decision tree, calculation formula such as formula (7), if decision tree it
Between similarity be greater than 60%, then the test performance in comparison step 6.3, retains the decision tree done very well, and forms random gloomy
Woods.
DT in formula (7)1With DT2Indicate two decision trees of progress similarity calculation；K indicates DT1With DT2To test case
The identical number of classification results；r1nWith r2nWhen indicating that n-th classification results are identical, DT1With DT2The characteristic attribute used, c table
Show classification results；Work as r1n=r2nWhen, i.e. DT1With DT2When obtaining identical classification results with identical characteristic attribute, I (r1n.c,
r2nIt .c)=1, is otherwise the number of test case for 0, Nt.
Step 6.5: by Bayes's voting mechanism, calculating separately the weight w of every decision tree in T1 and T2, h is calculated public
Formula such as formula (8), (9) obtain forest scheduling scheduling rule 1 and random forest scheduling rule 2.
V represents the number that this decision tree correctly classifies to test case in formula (8), (9)；M is indicated to test case mistake
Classification number；
Step 7, scheduling rule use: instructing flexible job shop dynamic to adjust by excavated random forest scheduling rule
Degree.It specifically includes:
Step 7.1 selects workpieces processing problem according to the problem of solution for workpiece machine choice or idle machine,
It finds and random forest scheduling rule 1 corresponding to cluster belonging to the bumpy weather of current flexible job shop or random forest tune
Metric then 2.
Step 7.2, according to selected random forest scheduling rule, optimal method is selected by comparing two-by-two, in candidate machine
Most suitable workpiece or machine are selected in device set M or candidate artifacts set J.
Step 7.2.1, for workpiece machine choice problem, if m1, m2 are two machines in M, according to selected by step 7.1
The selection result of every decision tree in random forest scheduling rule is calculated in random forest scheduling rule 1, wraps in these results
Selection 1 and selection 2 have been contained (selecting 1 to represent, the former m1 is suitable, and it is suitable that selection 2 represents the latter m2).Work is selected for idle machine
Part problem, if j1, j2 are two workpiece in J, the random forest scheduling rule 2 according to selected by step 7.1 is calculated random
The selection result of every decision tree in forest scheduling rule, decision 1 and decision 2 are contained in these results, and (decision 1 represents the former
J1 is suitable, and it is suitable that decision 2 represents the latter j2).
Step 7.2.2: selection result WR, WR after every decision tree weighting are obtained by Bayes's voting mechanism and calculate public
Formula such as formula (10), and the average value AWR of weighted results is acquired, if AWR indicates that the former m1 or j1 is suitable less than 1.5, if AWR is greater than
1.5 indicate that the latter m2 or j2 are suitable.
WR=wC+hR (10)
C represents the classification results that this decision tree provides in formula (10)；R represents the equal of the classification results that all decision trees provide
Value, w, h calculation formula are shown in formula (8), (9).
Example: in certain scheduler task, needing workpieces processing JT1, JT2 ..., JT8 each 100, i.e., and 10 batches, they
Delivery date be respectively 20.0,22.0,14.0,21.0,19.0,22.0,18.0,23.0 process unit's times, the per pass of workpiece
Process time such as table one of the process on each machine.And mechanical disorder has occurred when the time is 4, in first of work of JT1
Find that its second operation work occurs to lack material after the completion of sequence, when the time is 10, the process time of workpiece all increases by 10%.
Each work pieces process timetable of table one
The scheduling scheme that is obtained by the flexible job shop dynamic dispatching method based on industrial big data as shown in figure 5,
Abscissa indicates the time in figure, and ordinate indicates machine, the percentile digital representation workpiece type in Gantt chart, unit numbers table
Show operation number.The Maximal Makespan of final scheme is 21.8 process unit's times, 5.3 process unit's time of Zong Tuoqi, machine
96.4 process unit's time of total load.
Flexible job shop dynamic scheduling problem can smoothly be solved using patented method, and excavated using this method
The property made is strong, computational efficiency is high to instruct flexible job shop scheduling to have actually for scheduling rule, without carrying out to scheduling problem
The features such as modeling, the disturbance in energy real-time response workshop.
Claims (1)
1. a kind of flexible job shop dynamic dispatching method based on industrial big data, comprising the following steps:
Step 1, data acquisition: using the metadata acquisition tool Sqoop and Flume under Hadoop ecosystem, from MES, ERP,
Historical data relevant to scheduling is acquired in the existing information system such as SCADA, and is stored in HDFS file system；Acquisition
Data include three parts Dh={ d1, d2, d3 }: d1 is system disturbance information relevant to disturbance when scheduling scheme is formulated；D2 is
When certain procedure of workpiece selects processing machine, the state of every machine in the collection of machines of this procedure can be currently processed
Information；When d3 is that idle machine needs to select workpiece to be processed in waiting list, the shape of each workpiece in current queue
State information；
Step 2, Data Integration: by Tool for Data Warehouse Hive using SQL statement to the scheduling data in HDFS file system
Set DhIt is divided by unit of scheduling scheme, i.e., the scheduling relevant historical data generated primary scheduling scheme in commission
It is divided into together；
Step 3, the d2 and d3 in data after integration: being partially converted into the form of trained example using Spark by data conversion,
Rule digging is scheduled convenient for data mining algorithm；It specifically includes:
Step 3.1: for the scheduling data acquisition system D of acquisitionhThe part d2, by the machine of actual selection in certain history scheduling scheme
M1 is considered as most suitable machine, and the machine in the alternative collection of machines { m2, m3... } of this process is can be processed one by one with other in it
Compare to form trained example；
Step 3.2: for the scheduling data acquisition system D of acquisitionhThe part d3, by the workpiece of actual selection in certain history scheduling scheme
J1 is considered as most suitable machine, by its with other etc. workpiece in workpiece set { j2, j3... } to be processed compare to be formed one by one
Training example；
Step 4, data screening: from three Maximal Makespan, total tardiness time, machine total load indexs to history dispatching party
Case is considered, and screening obtains showing the scheduling relevant historical data set D that good scheduling scheme generates in commissionb；Tool
Body includes:
Step 4.1: in Maximal Makespan index, to only use the scheduling scheme of SPT rule generation in situation of the same race
Maximal Makespan is as screening criteria；It only uses SPT rule and refers to that workpiece selection is processed most fast machine and idle machine and selected
Select process time shortest workpiece；The scheduling scheme that Maximal Makespan meets this index enters step 4.2, is unsatisfactory for, and washes in a pan
It eliminates；
Step 4.2: in total tardiness time index, to use EDD rule combination SPT rule to complete same case in situation of the same race
Under scheduler task total extension time as screening criteria；SPT+EDD rule refers to that most fast machine is processed in workpiece selection
Device and idle machine select delivery date earliest workpiece；The scheduling scheme that total tardiness time meets this index enters step 4.3, no
Satisfaction is then eliminated；
Step 4.3: in machine total load index, to combine LMWT and SPT rule to complete the machine of scheduler task in situation of the same race
For device total load as screening criteria, LMWT+SPT rule refers to workpiece selection free time longest machine and idle machine
Select process time shortest workpiece；The scheduling data set that the scheduling scheme of these three indexs generates in commission can be met simultaneously
It closes, it will the input as scheduling rule mining algorithm；
Step 5；Cluster based on disturbance attribute: using DBSCAN to DbUsing scheduling scheme as unit, i.e., a scheduling scheme produces
Raw data are as an object, according to DbIn solution formulation when system disturbance attribute, i.e. the part d1 is carried out based on disturbance
The cluster of attribute；It specifically includes:
Step 5.1: to d1 partial data standardization, if certain data of disturbance attribute in each scheme is X1, X2,
X3..., Xn, then they need to convert by such as formula (1)；
In formula (1)Indicate the mean value of attribute；S is expressed as standard deviation；Y1, Y2, Y3..., Yn are the data after standardization；
Step 5.2: the parameter field radius Eps of DBSCAN algorithm is determined, with pair included at least in the radius of kernel object field
As number MinPts；
Step 5.3: finding out the kernel object p of one not processed (not being classified as some cluster or labeled as noise), core at random
The object number for including in the radius of the field heart object p is not less than MinPts, establishes new cluster C, will be all in p radius of neighbourhood Eps
Candidate Set N is added in object；
Step 5.4: finding out not yet processed object q in a Candidate Set N at random；If q is kernel object, by q neighborhood half
Object that is not processed and not being added to N is added in N in diameter Eps；If q is not included into any one cluster, C is added in q；
Step 5.5: step 5.4 is repeated, until N is sky；
Step 5.6: repeating step 5.3,5.4,5.5, when not new object can be added to any cluster, the process knot
Beam
Step 6, random forest scheduling rule are excavated: being used improved random forests algorithm, dug respectively from cluster each after cluster
The forest scheduling scheduling rule 1 that pick is resolved workpiece selection machine problem selects workpieces processing with idle machine is solved the problems, such as
Random forest scheduling 2；It specifically includes:
Step 6.1: for the cluster after each cluster, (being excavated random from d2 (excavating random forest scheduling rule 1) and the d3 in cluster
Forest scheduling rule 2) in extract trained example with putting back to, new training the example collection P1 and P2 of k are respectively formed, for constructing
Decision tree；
Step 6.2:P1 and P2 randomly chooses m characteristic attribute from d2 and d3 respectively, and calculates best divisional mode, instructs respectively
Get k decision tree T1 and T2；
Wherein decision tree building process are as follows:
Step 6.2.1: creation root node N；
Step 6.2.2: whether there are also remaining training examples for training of judgement example collection, if the return node N without if, if under having
One step；
Step 6.2.3: whether the scheduling decision of training of judgement example collection residue training example is all C, if then return node
N, and it is labeled as class C, if having in next step；
Step 6.2.4: judge to produce attribute list whether be it is empty, it is empty then be labeled as most classes occur in sample, it is otherwise next
Step；
Step 6.2.5: check whether the attribute in Attribute class table is continuity, and connection attribute will obtain attribute by dichotomy
The maximum attribute separate mode of gain G (D, A)；(all properties value of attribute can be divided by two parts by dichotomy, this
Shared N-1 kind division methods, the division threshold value of dichotomy are selection two respectively in the average value of adjacent two o'clock；Information gain calculating side
The following formula of formula (2), (3), (4)；
G (D, A)=H (D)-H (D | A) (2)
G (D, A) indicates the information gain of attribute A in formula (2)；H (D, A) classification information entropy in formula (3)；H in formula (4) (D | A) table
Show conditional entropy；In addition, D indicates training Exemplar Data Set, | D | indicate the training example quantity of D, and D has K classification Ck, k=
1,2；|Ck| it indicates in classification CkIn training example number；D can be divided into n subset D by attribute A1, D2...,
Dn, | Di| it is DiTraining example number；DiIn belong to class CkThe collection of training example be combined into Dik, | Dik| it is DikTraining example
Number；
Step 6.2.6: the selection maximum attribute flag node N of information gain-ratio, information gain-ratio calculation formula such as formula (5),
(6), return step 6.2.2；
GR (D, A)=G (D, A)/H (A) (5)
GR (D, A) indicates information gain-ratio in formula (5)；H (A) indicates division information；Other symbol meanings are same as above；
Step 6.3: using non-selected trained example in d2 and d3, the classification for testing decision tree in T1 and T2 respectively is showed；
Step 6.4: calculating the similarity S in T1 or T2 between decision tree, calculation formula such as formula (7), if between decision tree
Similarity is greater than 60%, then the test performance in comparison step 6.3, retains the decision tree done very well, form random forest；
DT in formula (7)1With DT2Indicate two decision trees of progress similarity calculation；K indicates DT1With DT2Classify to test case and ties
The identical number of fruit；r1nWith r2nWhen indicating that n-th classification results are identical, DT1With DT2The characteristic attribute used, c presentation class knot
Fruit；Work as r1n=r2nWhen, i.e. DT1With DT2When obtaining identical classification results with identical characteristic attribute, I (r1n.c,r2n.c)=
1, it is otherwise the number of test case for 0, Nt；
Step 6.5: by Bayes's voting mechanism, calculating separately the weight w of every decision tree in T1 and T2, h, calculation formula is such as
Formula (8), (9) obtain forest scheduling scheduling rule 1 and random forest scheduling rule 2；
V represents the number that this decision tree correctly classifies to test case in formula (8), (9)；M indicates to classify to test case mistake
Number；
Step 7, scheduling rule use: instructing flexible job shop dynamic dispatching by excavated random forest scheduling rule；Tool
Body includes:
Step 7.1 is found according to being that workpiece machine choice or idle machine select workpieces processing problem the problem of solution
Random forest scheduling rule 1 corresponding to cluster belonging to bumpy weather with current flexible job shop or random forest scheduling rule
Then 2；
Step 7.2, according to selected random forest scheduling rule, optimal method is selected by comparing two-by-two, in candidate machine collection
It closes in M or candidate artifacts set J and selects most suitable workpiece or machine；
Step 7.2.1, it is random according to selected by step 7.1 if m1, m2 are two machines in M for workpiece machine choice problem
The selection result of every decision tree in random forest scheduling rule is calculated in forest scheduling rule 1, contains in these results
Selection 1 and selection 2, selecting 1 to represent, the former m1 is suitable, and it is suitable that selection 2 represents the latter m2；Idle machine selection workpiece is asked
Topic, if j1, j2 are two workpiece in J, random forest scheduling rule 2, is calculated random forest according to selected by step 7.1
The selection result of every decision tree in scheduling rule, decision 1 and decision 2 are contained in these results, and decision 1 represents the former j1 and closes
Suitable, it is suitable that decision 2 represents the latter j2；
Step 7.2.2: selection result WR, the WR calculation formula after every decision tree weighting are obtained such as by Bayes's voting mechanism
Formula (10), and the average value AWR of weighted results is acquired, if AWR indicates that the former m1 or j1 is suitable less than 1.5, if AWR is greater than 1.5
Indicate that the latter m2 or j2 are suitable；
WR=wC+hR (10)
C represents the classification results that this decision tree provides in formula (10)；R represents the mean value for the classification results that all decision trees provide,
W, h calculation formula are shown in formula (8), (9).
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910144370.5A CN109902954B (en) | 2019-02-27 | 2019-02-27 | Flexible job shop dynamic scheduling method based on industrial big data |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910144370.5A CN109902954B (en) | 2019-02-27 | 2019-02-27 | Flexible job shop dynamic scheduling method based on industrial big data |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109902954A true CN109902954A (en) | 2019-06-18 |
| CN109902954B CN109902954B (en) | 2020-11-13 |
Family
ID=66945563
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910144370.5A Active CN109902954B (en) | 2019-02-27 | 2019-02-27 | Flexible job shop dynamic scheduling method based on industrial big data |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109902954B (en) |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111047215A (en) * | 2019-12-09 | 2020-04-21 | 中国兵器科学研究院 | Random forest based field replaceable unit classification and classification determination method |
| CN111401427A (en) * | 2020-03-12 | 2020-07-10 | 华中科技大学 | Product cost evaluation method and system based on industrial big data |
| CN111766839A (en) * | 2020-05-09 | 2020-10-13 | 同济大学 | Computer implementation system for self-adaptive updating of intelligent workshop scheduling knowledge |
| CN112712289A (en) * | 2021-01-18 | 2021-04-27 | 上海交通大学 | Adaptive method, system, and medium based on temporal information entropy |
| CN112883640A (en) * | 2021-02-04 | 2021-06-01 | 西南交通大学 | Digital twin station system, job scheduling method based on system and application |
| CN112904818A (en) * | 2021-01-19 | 2021-06-04 | 东华大学 | Prediction-reaction type scheduling method for complex structural member processing workshop |
| WO2021189620A1 (en) * | 2020-03-25 | 2021-09-30 | 重庆邮电大学 | Digital workshop electric energy management research method based on context awareness |
| CN115357570A (en) * | 2022-08-24 | 2022-11-18 | 安徽维德工业自动化有限公司 | Workshop optimization scheduling management method based on random forest algorithm |
| CN116402173A (en) * | 2022-09-06 | 2023-07-07 | 大连理工大学 | Intelligent algorithm for distributing container areas and container positions of ship unloading container based on machine learning |
| CN117010671A (en) * | 2023-10-07 | 2023-11-07 | 中国信息通信研究院 | Distributed flexible workshop scheduling method and device based on block chain |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104914835A (en) * | 2015-05-22 | 2015-09-16 | 齐鲁工业大学 | Flexible job-shop scheduling multi-objective method |
| CN106094757A (en) * | 2016-07-15 | 2016-11-09 | 郑州航空工业管理学院 | A kind of dynamic flexible solving job shop scheduling problem control method based on data-driven |
| CN106611232A (en) * | 2016-02-04 | 2017-05-03 | 四川用联信息技术有限公司 | Layered optimization algorithm for solving multi-technical-route workshop scheduling |
| CN107862411A (en) * | 2017-11-09 | 2018-03-30 | 西南交通大学 | A kind of extensive flexible job shop scheduling optimization method |
| CN108733003A (en) * | 2017-04-20 | 2018-11-02 | 南京理工大学 | Slewing parts process working hour prediction technique based on kmeans clustering algorithms and system |
-
2019
- 2019-02-27 CN CN201910144370.5A patent/CN109902954B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104914835A (en) * | 2015-05-22 | 2015-09-16 | 齐鲁工业大学 | Flexible job-shop scheduling multi-objective method |
| CN106611232A (en) * | 2016-02-04 | 2017-05-03 | 四川用联信息技术有限公司 | Layered optimization algorithm for solving multi-technical-route workshop scheduling |
| CN106094757A (en) * | 2016-07-15 | 2016-11-09 | 郑州航空工业管理学院 | A kind of dynamic flexible solving job shop scheduling problem control method based on data-driven |
| CN108733003A (en) * | 2017-04-20 | 2018-11-02 | 南京理工大学 | Slewing parts process working hour prediction technique based on kmeans clustering algorithms and system |
| CN107862411A (en) * | 2017-11-09 | 2018-03-30 | 西南交通大学 | A kind of extensive flexible job shop scheduling optimization method |
Non-Patent Citations (2)
| Title |
|---|
| THOMAS BUSCHMANN ET AL: "《Flexible and Robust Walking(Workshop-Dynamic Locomation and Balancing of Humanoids: State of the Art and Challenges)》", 《2015 IEEE INTERNATIONAL CONFERENCE ON ROBOTICS AND AUTOMATION》 * |
| 汪双喜等: "《不同再调度周期下的柔性作业车间动态调度》", 《万方数据》 * |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111047215B (en) * | 2019-12-09 | 2023-06-23 | 中国兵器科学研究院 | Method for determining classification of field replaceable units based on random forest |
| CN111047215A (en) * | 2019-12-09 | 2020-04-21 | 中国兵器科学研究院 | Random forest based field replaceable unit classification and classification determination method |
| CN111401427A (en) * | 2020-03-12 | 2020-07-10 | 华中科技大学 | Product cost evaluation method and system based on industrial big data |
| CN111401427B (en) * | 2020-03-12 | 2022-11-08 | 华中科技大学 | Product cost evaluation method and system based on industrial big data |
| WO2021189620A1 (en) * | 2020-03-25 | 2021-09-30 | 重庆邮电大学 | Digital workshop electric energy management research method based on context awareness |
| CN111766839A (en) * | 2020-05-09 | 2020-10-13 | 同济大学 | Computer implementation system for self-adaptive updating of intelligent workshop scheduling knowledge |
| CN111766839B (en) * | 2020-05-09 | 2023-08-29 | 同济大学 | Computer-implemented system for self-adaptive update of intelligent workshop scheduling knowledge |
| CN112712289A (en) * | 2021-01-18 | 2021-04-27 | 上海交通大学 | Adaptive method, system, and medium based on temporal information entropy |
| CN112712289B (en) * | 2021-01-18 | 2022-11-22 | 上海交通大学 | Adaptive method, system, and medium based on temporal information entropy |
| CN112904818A (en) * | 2021-01-19 | 2021-06-04 | 东华大学 | Prediction-reaction type scheduling method for complex structural member processing workshop |
| CN112904818B (en) * | 2021-01-19 | 2022-07-15 | 东华大学 | Prediction-reaction type scheduling method for complex structural member processing workshop |
| CN112883640A (en) * | 2021-02-04 | 2021-06-01 | 西南交通大学 | Digital twin station system, job scheduling method based on system and application |
| CN115357570A (en) * | 2022-08-24 | 2022-11-18 | 安徽维德工业自动化有限公司 | Workshop optimization scheduling management method based on random forest algorithm |
| CN116402173A (en) * | 2022-09-06 | 2023-07-07 | 大连理工大学 | Intelligent algorithm for distributing container areas and container positions of ship unloading container based on machine learning |
| CN117010671A (en) * | 2023-10-07 | 2023-11-07 | 中国信息通信研究院 | Distributed flexible workshop scheduling method and device based on block chain |
| CN117010671B (en) * | 2023-10-07 | 2023-12-05 | 中国信息通信研究院 | Distributed flexible workshop scheduling method and device based on block chain |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109902954B (en) | 2020-11-13 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109902954A (en) | A kind of flexible job shop dynamic dispatching method based on industrial big data | |
| KR102184182B1 (en) | Project/Task Intelligent Goal Management Method and Platform based on Super Tree | |
| Shen et al. | Mathematical modeling and multi-objective evolutionary algorithms applied to dynamic flexible job shop scheduling problems | |
| Guo et al. | Optimisation of integrated process planning and scheduling using a particle swarm optimisation approach | |
| US20210073695A1 (en) | Production scheduling system and method | |
| CN113792924A (en) | Single-piece job shop scheduling method based on Deep reinforcement learning of Deep Q-network | |
| CN103310285A (en) | Performance prediction method applicable to dynamic scheduling for semiconductor production line | |
| Li et al. | Data-based scheduling framework and adaptive dispatching rule of complex manufacturing systems | |
| He et al. | Integrated scheduling of production and distribution operations in a global MTO supply chain | |
| Pani et al. | A data mining approach to forecast late arrivals in a transhipment container terminal | |
| Zhao et al. | An improved Q-learning based rescheduling method for flexible job-shops with machine failures | |
| CN102402716A (en) | Intelligent production decision support system | |
| CN104572297A (en) | Hadoop job scheduling method based on genetic algorithm | |
| CN109039727A (en) | Message queue monitoring method and device based on deep learning | |
| CN106327053B (en) | Construction method of weaving process recommendation model based on multi-mode set | |
| Ajorlou et al. | Optimization of a multiproduct conwip-based manufacturing system using artificial bee colony approach | |
| CN111626497A (en) | People flow prediction method, device, equipment and storage medium | |
| Ishankhodjayev et al. | Development of information support for decision-making in intelligent energy systems | |
| CN107463151B (en) | A kind of complex surface machining multidimensional knowledge cloud cooperating service method | |
| Chan et al. | The applications of flexible manufacturing technologies in business process reengineering | |
| CN115689201A (en) | Multi-criterion intelligent decision optimization method and system for enterprise resource supply and demand allocation | |
| Shen et al. | Blocking flow shop scheduling based on hybrid ant colony optimization | |
| CN105260948B (en) | A kind of water-supply systems daily planning scheduling decision method | |
| Madureira et al. | Using genetic algorithms for dynamic scheduling | |
| Aksvonov et al. | Development of a hybrid decision-making method based on a simulation-genetic algorithm in a web-oriented metallurgical enterprise information system |
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 |