CN110866591B - Method for carrying out prospective cloud manufacturing service lease configuration based on demand prediction - Google Patents

Method for carrying out prospective cloud manufacturing service lease configuration based on demand prediction Download PDF

Info

Publication number
CN110866591B
CN110866591B CN201911030698.0A CN201911030698A CN110866591B CN 110866591 B CN110866591 B CN 110866591B CN 201911030698 A CN201911030698 A CN 201911030698A CN 110866591 B CN110866591 B CN 110866591B
Authority
CN
China
Prior art keywords
service
manufacturing
time
decision
task
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN201911030698.0A
Other languages
Chinese (zh)
Other versions
CN110866591A (en
Inventor
方水良
陈晟恺
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Zhejiang University ZJU
Original Assignee
Zhejiang University ZJU
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Zhejiang University ZJU filed Critical Zhejiang University ZJU
Priority to CN201911030698.0A priority Critical patent/CN110866591B/en
Publication of CN110866591A publication Critical patent/CN110866591A/en
Application granted granted Critical
Publication of CN110866591B publication Critical patent/CN110866591B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0201—Market modelling; Market analysis; Collecting market data
    • G06Q30/0202—Market predictions or forecasting for commercial activities
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/06—Buying, selling or leasing transactions
    • G06Q30/0645—Rental transactions; Leasing transactions
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/04—Manufacturing
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
    • Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
    • Y02P90/30—Computing systems specially adapted for manufacturing

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Accounting & Taxation (AREA)
  • General Physics & Mathematics (AREA)
  • Strategic Management (AREA)
  • Finance (AREA)
  • Data Mining & Analysis (AREA)
  • Development Economics (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Mathematical Physics (AREA)
  • Marketing (AREA)
  • Computational Linguistics (AREA)
  • Evolutionary Computation (AREA)
  • Software Systems (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Computing Systems (AREA)
  • Economics (AREA)
  • General Engineering & Computer Science (AREA)
  • Molecular Biology (AREA)
  • General Business, Economics & Management (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Manufacturing & Machinery (AREA)
  • Human Resources & Organizations (AREA)
  • Primary Health Care (AREA)
  • Tourism & Hospitality (AREA)
  • Game Theory and Decision Science (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses a method for carrying out prospective cloud manufacturing service lease configuration based on demand prediction, which comprises the following innovation contents: 1) Formally constructing a model of manufacturing service and manufacturing task and a cloud service mode thereof; 2) Providing an analysis method of the user task order, and classifying based on task characteristics; 3) Triggering online scheduling by a set order number threshold value based on actual order demands and virtual predicted order demands; 4) Predicting the demand of the short-term virtual order based on an LSTM model, and combining the demand with an actual order to perform optimized scheduling; 5) And designing a corresponding optimized scheduling engine, and determining the resource service optimized lease configuration of the virtual and real order tasks of the scheduling window. The method has the specific effects of fast order demand response of the user, short enterprise decision time and high comprehensive quality of cloud platform service matching and leasing.

Description

Method for carrying out prospective cloud manufacturing service lease configuration based on demand prediction
Technical Field
The invention relates to a cloud manufacturing service leasing method, in particular to a method for scheduling manufacturing tasks by combining an LSTM (Long Short Term Memory) model to carry out demand prediction, so as to further configure cloud manufacturing service leasing.
Background
As an advanced manufacturing mode oriented to manufacturing services, cloud manufacturing is based on technologies such as information technology, internet of things, virtualization and cloud computing, and achieves virtualization and service of manufacturing resources. The service demander can apply for using the cloud-end manufacturing service at any time according to the requirement, and describe the requirement as a manufacturing task to be distributed on the platform; service providers share their manufacturing resources and manufacturing capabilities in the form of services through the virtualization encapsulation techniques provided by the platform.
As the complexity of cloud manufacturing requirements increases, simple manufacturing service configuration rentals have failed to meet the requirements of customers in terms of construction period, quality, cost, etc., and cloud platforms need to better meet the requirements of a large number of users through advanced service rental configuration methods.
The cloud manufacturing platform collects scattered manufacturing services, and the user is required to use the corresponding services in a renting mode through related modules. In a cloud manufacturing environment, lease configuration of a manufacturing service refers to a process of mapping a manufacturing task requirement to a corresponding manufacturing service for processing. The processing of the manufacturing task requires a certain amount of time and capacity to be occupied on specific service resources, and the different manufacturing service providers have different capacities of service resources and different lease states thereof, which results in different capacities of manufacturing tasks that can be assumed by the different manufacturing service providers. Thus, the cloud platform needs to dynamically optimize the orchestration rental service to meet the needs of numerous manufacturing tasks. The reasonable service lease configuration not only meets the requirements of a demand side on minimizing lease cost, maximizing service use effectiveness, minimizing task delay value and the like, but also considers the uncertainty of service available capacity.
Disclosure of Invention
The essence of the reasonable lease service problem on the cloud platform is a comprehensive optimization scheduling problem of a plurality of groups of parallel resources, generally, a workload-based analysis method can be adopted, a corresponding optimization model is established to solve by taking service utility, service quality, service cost and the like as targets, and then each task is optimized and arranged to a corresponding resource point according to a result. The method has higher real-time requirement, and the solution of the method depends on a solution algorithm capable of responding quickly, such as a scheduling result prediction method based on improved learning rate. For a single customer, the irregular generation of demand is a mode of cloud platform manufacturing task arrival, such as periodic ordering plans and the like. This is an extended resource constrained scheduling problem, subject to the capacity available for the manufacturing service. However, in a cloud manufacturing environment, the uncertainty of different task requirements of a plurality of clients and the diversity of task structures of the clients increase the decision difficulty of the problem, and the traditional global optimization model and the corresponding algorithm are not applicable any more.
The invention aims to provide efficient and quick cloud manufacturing service lease configuration so as to improve the resource utilization rate and the satisfaction degree of a user. However, at present, cloud manufacturing services do not have a satisfactory service model and a corresponding usage mode, and it is difficult to apply and verify the effect.
Determining the service lease configuration of an order by a cloud manufacturing platform faces the following technical problem in 4 respects:
1) Modeling a cloud manufacturing service mode, wherein the modeling comprises the expression of manufacturing tasks and manufacturing resources and service lease configuration;
2) A trigger method of a service lease configuration decision maker;
3) A training method of a demand prediction model for assisting a lease decision;
4) A method for on-line task scheduling.
In order to solve the technical problems, the technical scheme adopted by the invention is as follows:
a method for carrying out prospective cloud manufacturing service lease configuration based on demand prediction comprises the following steps:
(1) Building a model of manufacturing services and manufacturing tasks;
(2) Analyzing the manufacturing tasks released by the users, classifying the released manufacturing tasks according to the service types required by the manufacturing tasks, and placing the manufacturing tasks into waiting queues of the same type of manufacturing services
Figure GDA0003816047040000021
Aiming at the triggering method of the service lease configuration decision maker, the invention adopts a double judgment criterion based on the waiting queue capacity and the waiting interval duration as a decision triggering method. Specifically, as shown in fig. 4, the decision-maker will enable the sub-routine of listening to the arriving MT, and the arriving MT will enter the corresponding MS according to the type of MS it needsWait queue
Figure GDA0003816047040000022
If the decision time of the previous time is tlastThe current time is tcurrentIf t is satisfiedcurrent-tlastIf the time is more than or equal to delta T, immediately executing a scheduling decision, wherein the delta T represents a set time interval threshold; when a decision is triggered, if the queue is currently waiting
Figure GDA0003816047040000023
Capacity of (2)
Figure GDA0003816047040000024
Then, a scheduling decision is immediately made, i.e. go to step (4), where NτIndicating a queue capacity threshold, and if a decision is triggered,
Figure GDA0003816047040000025
turning to the step (3);
(3) Calling a trained LSTM-based demand forecasting module to generate short-term virtual order demands
Figure GDA0003816047040000026
Merging with waiting queue into scheduling decision queue
Figure GDA0003816047040000027
(4) And the decision maker determines the lease configuration of each task in the queue through the scheduling engine and summarizes and outputs the lease configuration.
Specifically, the method comprises the following steps:
(1) And aiming at cloud manufacturing service mode modeling, mathematical models of manufacturing tasks and manufacturing services are respectively established. A manufacturing task is a manifestation of customer demand and requires a certain amount of time and capacity to be handled by the corresponding manufacturing service. The manufacturing task of customer j issued to the platform s-th time is marked as MT (j, s), and the complete attribute composition is shown as the tuple of formula (1), including the submission time
Figure GDA0003816047040000028
Construction period
Figure GDA0003816047040000029
Required capacity
Figure GDA00038160470400000210
Type of service required
Figure GDA00038160470400000211
Required processing time pjAnd the like. For a particular MT (j, s), the values of the attributes in the tuple are static.
Figure GDA00038160470400000212
Wherein the content of the first and second substances,
Figure GDA00038160470400000213
representing a set of customer numbers. The manufacturing service is a task processing unit of the cloud manufacturing platform and is formed by virtualizing and packaging manufacturing resource capacities from different enterprises. The manufacturing service registered by the manufacturing service provider i on the platform is denoted as MS (i), and is composed of the maximum rentable capacity KiService type phiiUnit rental fee CiService utility factor PiWaiting to process task queue Q at time ti(t) and set of tasks being processed
Figure GDA00038160470400000214
And (3) the tuple composed of the equal attributes is represented as formula (2). Except for Qi(t) and
Figure GDA00038160470400000215
except for the time variable, the other attributes are static values.
Figure GDA0003816047040000031
Wherein the content of the first and second substances,
Figure GDA0003816047040000032
representing a set of service provider numbers. Taking the MT with the number j issued to the platform s for the second time as an example, the manufacturing service lease configuration refers to the cloud platform determining the MS (i) used by the MT (j, s) processing and the start-stop time and the service occupation amount thereof. Satisfying type matching conditions according to MT (j, s) configuration
Figure GDA0003816047040000033
MS (i), i.e. at the start-stop time [ b ]j,i,ej,i) And the interval occupies part of the service capacity of the MS (i), and meets the type matching and processing time length constraints (3) - (4).
Figure GDA0003816047040000034
Figure GDA0003816047040000035
Wherein, the first and the second end of the pipe are connected with each other,
Figure GDA0003816047040000036
indicating a predicate function, i.e., a value of 1 when the condition is satisfied and 0 otherwise.
(3) Aiming at the training method of the demand prediction model for assisting the leasing decision, the invention adopts a method of moving a window to normalize the input and output patterns required by the LSTM model. Specifically, taking customer j as an example, the platform collects MT related data periodically published to the platform, and takes the number of publication times s as a step amount, and considers the corresponding MS demand capacity as sequence data listed in equation (5).
Figure GDA0003816047040000037
In order to import the sampled sequence data into the LSTM model, data transformation by a moving window method is required. The sequence data is moved to the right by w time units each time by taking a window with a fixed size as a unit, and input original data is integrated into (6) to be used as an input and output data unit of the model.
Figure GDA0003816047040000038
Where t represents a time window sequence number. Because the prediction problem in the invention does not belong to the classification problem, and the outputs have the same importance degree, the mean square error is used as a loss function of the prediction model training, and the RMSProp optimization algorithm is used for optimizing the gradient descent process.
According to the invention, the manufacturing service capacity demand is predicted by collecting service capacity demand historical data and building an LSTM prediction model based on Tensorflow 1.2.1, and the trained LSTM model can be used for demand prediction.
(4) For the method of efficient online application scheduling, the present invention first determines the optimization objectives in terms of 3, respectively representing cost, service utility, task delay, etc., as shown in (7) - (9). Wherein, the first and the second end of the pipe are connected with each other,
Figure GDA0003816047040000041
and
Figure GDA0003816047040000042
the lease cost, service utility and delay value of the MT (j, s) are respectively expressed.
Figure GDA0003816047040000043
Figure GDA0003816047040000044
Figure GDA0003816047040000045
For each MT (j, s), the service lease cost can be expressed as equation (10), i.e., the product of the lease duration, capacity and its unit price of the selected MS.
Figure GDA0003816047040000046
Obtaining quality services, similar to service rental costs, is also an important optimization goal to attract customers. For each MT, the usage utility of the MS after completion thereof can be expressed as formula (11), i.e., the usage utility of the selected MS.
Figure GDA0003816047040000047
Lower task latency is an important indicator of meeting customer demand, the value of which is related to the outcome of the service lease configuration. For each MT, the overdue amount is a time interval beyond the contracted period, and can be represented by (12).
Figure GDA0003816047040000048
Decision vector y(τ)By aggregation of decision task queues
Figure GDA0003816047040000049
And (5) a decision variable (14) related to the MT, which represents the starting processing time of the MT (j, s).
Figure GDA00038160470400000410
Figure GDA0003816047040000051
Wherein the content of the first and second substances,
Figure GDA0003816047040000052
indicating that MT (j, s) selects the decision variable to process on MS (i). As such, for a set of decision task queues
Figure GDA0003816047040000053
The optimization objective for the manufacturing service rental configuration may be represented as (15).
Figure GDA0003816047040000054
The method is a multi-objective optimization problem, needs to consider factors in 3 aspects at the same time, can establish a constraint planning model, and calls a mature solver to solve. Obtained by solving
Figure GDA0003816047040000055
And required capacity for attributes of MT
Figure GDA0003816047040000056
I.e. the rental configuration of the service. Can be invoked by
Figure GDA0003816047040000057
The CP Optimizer 12.7.0.0 engine does the solution of the service lease configuration.
Compared with the prior art, the invention has the following effects:
(1) The cloud manufacturing service mode is normalized, a general modeling method is adopted to model a manufacturing task, a manufacturing service and a trigger flow, and practical application of the manufacturing service is realized;
(2) Through demand forecasting, service providing enterprises can pre-judge the short-term future resource capacity condition and perform comprehensive optimized resource scheduling;
(3) The event-driven manufacturing service lease configuration method provides a quick decision method for each manufacturing service provider, can reduce the service lease cost on the whole, improves the service use utility and reduces the delay rate of the undertaken tasks.
Drawings
FIG. 1 is a system framework consisting of cloud manufacturing service rental function modules based on the LSTM model;
FIG. 2 manufacturing service rental schematic;
FIG. 3 is a schematic of the moving window used in the LSTM model;
FIG. 4 decision flow of the decision maker;
FIG. 5 selection of parameters for the demand module.
Detailed Description
The technical scheme of the invention is further explained by combining the attached drawings.
The method for carrying out prospective cloud manufacturing service lease configuration based on demand prediction specifically comprises the following steps:
and aiming at cloud manufacturing service mode modeling, mathematical models of manufacturing tasks and manufacturing services are respectively established.
A manufacturing task is a manifestation of customer demand and requires a certain amount of time and capacity to be handled by the corresponding manufacturing service. The manufacturing task of customer j issued to the platform s-th time is marked as MT (j, s), and the complete attribute composition is shown as the tuple of formula (1), including the submission time
Figure GDA0003816047040000058
Construction period
Figure GDA0003816047040000059
Required capacity
Figure GDA00038160470400000510
Type of service required
Figure GDA00038160470400000511
Required processing time pjAnd the like. For a particular MT (j, s), the values of the attributes in the tuple are static.
Figure GDA0003816047040000061
Wherein the content of the first and second substances,
Figure GDA0003816047040000062
representing a set of customer numbers. The manufacturing service is a task processing unit of the cloud manufacturing platform and is from different enterprisesThe manufacturing resource capability is formed by virtualization and packaging. The manufacturing service registered by the manufacturing service provider i on the platform is denoted as MS (i), and is composed of the maximum rentable capacity KiService type phiiUnit rental fee CiService utility factor PiWaiting for processing task queue Q at time ti(t) and set of tasks being processed
Figure GDA0003816047040000063
And the tuple represented by the formula (2) is formed by equal attributes. Except for Qi(t) and
Figure GDA0003816047040000064
except for the time variable, the other attributes are static values.
Figure GDA0003816047040000065
Wherein the content of the first and second substances,
Figure GDA0003816047040000066
representing a set of service provider numbers. Taking the MT with the number j issued to the platform s for the second time as an example, the manufacturing service lease configuration refers to the cloud platform determining the MS (i) used by the MT (j, s) processing and the start-stop time and the service occupation amount thereof. Satisfying type matching conditions according to MT (j, s) configuration
Figure GDA0003816047040000067
MS (i), i.e. at the start-stop time [ b ]j,i,ej,i) And the interval occupies part of the service capacity of the MS (i), and meets the type matching and processing time length constraints (3) - (4).
Figure GDA0003816047040000068
Figure GDA0003816047040000069
Wherein the content of the first and second substances,
Figure GDA00038160470400000610
representing a predicate function, i.e., a value of 1 when the condition is satisfied, and 0 otherwise.
Aiming at the triggering method of the service lease configuration decision maker, the invention adopts a double judgment criterion based on the waiting queue capacity and the waiting interval duration as a decision triggering method. Specifically, as shown in fig. 4, the decision device enables the sub-process of listening to the arriving task, and the arriving MT enters the corresponding waiting queue according to the MS type required by the MT
Figure GDA00038160470400000611
If the decision time of the previous time is tlastThe current time is tcurrentIf t is satisfiedcurrent-tlastIf the time is more than or equal to delta T, immediately executing a scheduling decision, wherein the delta T represents a set time interval threshold; if the current waiting queue
Figure GDA00038160470400000612
Capacity of (2)
Figure GDA00038160470400000613
Then a scheduling decision is immediately executed, where NτIndicating a queue capacity threshold. Wherein, if the decision is triggered,
Figure GDA00038160470400000614
then the trained demand forecasting model needs to be used to generate short-term virtual order demands
Figure GDA0003816047040000071
And combined with the waiting queue into a decision task queue set
Figure GDA0003816047040000072
Aiming at the training method of the demand prediction model for assisting the leasing decision, the invention adopts a method of moving a window to normalize the input and output patterns required by the LSTM model. Specifically, taking customer j as an example, the platform collects MT-related data periodically distributed to the platform, and takes the distribution number s as a step amount, and considers the corresponding MS demand capacity as sequence data listed in formula (5).
Figure GDA0003816047040000073
In order to import the sampled sequence data into the LSTM model, data transformation by a moving window method is required. The sequence data is shifted to the right by w time units each time by taking a window with a fixed size as a unit, and input original data is integrated into (6) to be used as an input/output data unit of the model.
Figure GDA0003816047040000074
Where t represents the time window number. Since the prediction problem in the present invention does not belong to the classification problem, and the outputs have the same importance, the mean square error is used as the loss function of the prediction model training, and the RMSProp optimization algorithm is used to optimize the gradient descent process.
According to the invention, the manufacturing service capacity demand is predicted by collecting service capacity demand historical data and building an LSTM prediction model based on Tensorflow 1.2.1, and the trained LSTM model can be used for demand prediction.
For the method of efficient online application scheduling, the present invention first determines the optimization objectives in terms of 3, respectively representing cost, service utility, task delay, etc., as shown in (7) - (9). Wherein the content of the first and second substances,
Figure GDA0003816047040000075
and
Figure GDA0003816047040000076
the lease cost, service utility and delay value of the MT (j, s) are respectively expressed.
Figure GDA0003816047040000077
Figure GDA0003816047040000078
Figure GDA0003816047040000079
For each MT (j, s), the service lease cost can be expressed as equation (10), i.e., the product of the lease duration, capacity and its unit price of the selected MS.
Figure GDA0003816047040000081
Obtaining quality services, similar to service rental costs, is also an important optimization goal to attract customers. For each MT, the usage utility of the MS after completion thereof can be expressed as formula (11), i.e., the usage utility of the selected MS.
Figure GDA0003816047040000082
Lower task latency is an important indicator of meeting customer demand, and its value is related to the outcome of the service lease configuration. For each MT, its overdue amount is a time interval beyond the contracted period, and can be represented by (12).
Figure GDA0003816047040000083
Decision vector y(τ)By aggregation of decision task queues
Figure GDA0003816047040000084
And a decision variable (14) related to the MT in the item (m) represents the processing starting time of the MT (j, s).
Figure GDA0003816047040000085
Figure GDA0003816047040000086
Wherein the content of the first and second substances,
Figure GDA0003816047040000087
indicating that MT (j, s) selects the decision variable to process on MS (i). As such, for a decision queue task set
Figure GDA0003816047040000088
The optimization objective for the manufacturing service rental configuration can be represented as (15).
Figure GDA0003816047040000089
The method is a multi-objective optimization problem, needs to consider factors in 3 aspects at the same time, can establish a constraint planning model, and calls a mature solver to solve. Obtained by solving
Figure GDA00038160470400000810
And required capacity for attributes of MT
Figure GDA00038160470400000811
I.e. the rental configuration of the service. Can be invoked by
Figure GDA00038160470400000812
The CP Optimizer 12.7.0.0 engine does the solution of the service lease configuration.
FIG. 1 is a system structure based on the method of the present invention, which mainly includes a task issuing system (c), a demand forecasting system (f) and a service lease configuration system (b) 3, and the final service lease configuration result is as shown in FIG. 2, i.e. the task occupies a certain duration and a certain capacity on the selected manufacturing service. Where the demand forecasting system needs to move through a window as shown in fig. 3 as input of sequence data and determine relevant parameters of the model through experiments as shown in fig. 5. The logic flow of the service lease configuration system is shown in fig. 4, wherein the decision-maker uses a module for real-time monitoring task arrival and determines whether to trigger the scheduling flow according to the threshold values of both queue length and time interval.
The real-time specific process of the invention is as follows:
(1) The task issuing system collects manufacturing tasks (a) regularly issued by a demand client in a task pool of the platform;
(2) Entering a waiting queue (d) of a corresponding service type according to the time sequence;
(3) The demand forecasting system forecasts the demands of different series of tasks, and merges the forecasting result and the tasks in the waiting queue into a scheduling decision queue (h);
(4) And (c) triggering the scheduling module (e) by the scheduler through the service lease configuration system, and calculating the service start-stop time (b) of each task in the waiting queue.
For example, the main steps are:
(1) The parameters of the platform are set as follows: the length of the decision queue is 6, and the interval time is 10;
(2) For both tasks MT (1, 1) and MT (1, 2) published to the platform, a type 2 manufacturing service needs to be used. After they are released to the platform, they enter the corresponding waiting queue;
(3) According to the current judgment of the decision maker, determining to carry out demand prediction on the corresponding manufacturing service so as to supplement 4 virtual manufacturing tasks to the corresponding decision queue;
(4) The output manufacturing service rental configuration is configured by invoking a scheduling decision maker integrated within the system to: the MT (1, 2) is scheduled for processing at the MS (1, 1) to a corresponding time window, and the MT (1, 1) is scheduled for processing at the MS (1, 2) to a corresponding time window.

Claims (1)

1. A method for carrying out prospective cloud manufacturing service lease configuration based on demand prediction comprises the following steps:
(1) Building a model of manufacturing services and manufacturing tasks; the method comprises the following specific steps:
(a) Analyzing the required characteristics of each manufacturing task and constructing a corresponding mathematical model
The manufacturing task of the customer j which is released to the platform for the s th time is recorded as MT (j, s), the complete attribute composition is shown as the tuple of the formula (1), and the submission time comprises
Figure FDA0003816047030000011
Construction period
Figure FDA0003816047030000012
Required capacity
Figure FDA0003816047030000013
Type of service required
Figure FDA0003816047030000014
Required processing duration pj(ii) a For a specific MT (j, s), the value of each attribute in the tuple is static;
Figure FDA0003816047030000015
wherein the content of the first and second substances,
Figure FDA0003816047030000016
representing a set of customer numbers;
(b) The manufacturing service is a task processing unit of the cloud manufacturing platform, and is formed by virtualizing and packaging manufacturing resource capacity from different enterprises
The manufacturing service registered by the manufacturing service provider i on the platform is denoted as MS (i), and is composed of the maximum rentable capacity KiService type phiiUnit rental fee CiService utility factor PiWaiting to process task queue Q at time ti(t) and set of tasks being processed
Figure FDA0003816047030000017
Attribute-composed tuple representation as shown in formula (2); except for Qi(t) and
Figure FDA0003816047030000018
except for the time variable, the other attributes are static values;
Figure FDA0003816047030000019
wherein the content of the first and second substances,
Figure FDA00038160470300000110
representing a set of service provider numbers;
(c) On the basis of models of tasks and services, corresponding service configuration models are constructed
Taking the MT with the serial number j issued to the platform s for the second time as an example, the manufacturing service lease configuration means that the cloud platform determines the MS (i) used by the MT (j, s) processing and the start-stop time and the service occupation amount thereof, and meets the type matching condition according to the MT (j, s) configuration
Figure FDA0003816047030000021
MS (i), i.e. at the start-stop time [ b ]j,i,ej,i) The interval occupies part of the service capacity of the MS (i), and meets the type matching and processing time length constraints (3) - (4);
Figure FDA0003816047030000022
Figure FDA0003816047030000023
wherein the content of the first and second substances,
Figure FDA0003816047030000024
representing a predicate function, namely the value of the predicate function is 1 when the condition is met, and otherwise, the value of the predicate function is 0;
(2) Analyzing the manufacturing tasks released by the users, classifying the released manufacturing tasks according to the service types required by the manufacturing tasks, and placing the manufacturing tasks into waiting queues of the same type of manufacturing services
Figure FDA0003816047030000025
Setting a scheduling trigger mechanism to determine whether to predict the demand:
(a) Adopting a dual threshold design based on time and the number of orders, considering the number of actually waiting tasks in each queue and considering whether the time interval between two decisions is too long or not; specifically, the time-based trigger indicates that the decision time of the previous time is tlastThe current time is tcurrentIf t is satisfiedcurrent-tlastWhen the time is more than or equal to delta T, immediately executing a scheduling decision, wherein the delta T represents a set time interval threshold; when a decision is triggered, if the queue is currently waiting
Figure FDA0003816047030000026
Capacity of (2)
Figure FDA0003816047030000027
Then, a scheduling decision is immediately executed, i.e. go to step (4), where N isτIndicating a queue capacity threshold, and if a decision is triggered,
Figure FDA0003816047030000028
turning to the step (3);
(b) A mode of circularly monitoring task arrival events is adopted, so that the system can effectively run; each arriving task detects whether a decision maker with double thresholds needs to be triggered or not, and then decides whether a decision for demand prediction needs to be made or not;
(3) Invoking a trained LSTM-based demand forecasting module to generate short-term virtual order demands
Figure FDA0003816047030000029
Merging with a waiting queue into a set of decision task queuesClosing box
Figure FDA00038160470300000210
(4) The decision maker determines the lease configuration of each task in the queue through a scheduling engine, and summarizes and outputs the lease configuration;
the training method of the LSTM-based demand prediction model in the step (3) is as follows:
(a) Converting the historical service demand capacity into sequence data, specifically, taking a client j as an example, acquiring MT (MT) related data periodically issued to a platform by the platform, taking the issuing times s as a stepping quantity, and regarding the corresponding MS demand capacity as the sequence data listed in a formula (5);
Figure FDA0003816047030000031
(b) Importing the sampled sequence data into an LSTM model, and performing data conversion by a moving window method; specifically, the sequence data moves to the right for w time units each time by taking a window with a fixed size as a unit, and input original data are integrated into (6) to be used as an input and output data unit of the model;
Figure FDA0003816047030000032
wherein t represents a time window sequence number; using the mean square error as a loss function of the prediction model training, and using a RMSProp optimization algorithm to optimize the gradient descent process;
the decision method for lease configuration in the step (4) is as follows:
(a) For the method of efficient online application scheduling, first determine the optimization objectives in terms of cost, service utility, and task deferral 3, respectively, as shown in (7) - (9), where,
Figure FDA0003816047030000033
and
Figure FDA0003816047030000034
lease cost, service utility and deferral value for MT (j, s), respectively:
Figure FDA0003816047030000035
Figure FDA0003816047030000036
Figure FDA0003816047030000037
wherein, the first and the second end of the pipe are connected with each other,
Figure FDA0003816047030000038
representing the decision task queue set, for each MT (j, s), the service lease cost can be expressed as equation (10), i.e., the product of lease duration, capacity and its unit price for the selected MS:
Figure FDA0003816047030000041
similar to the service lease cost, obtaining quality services is also an important optimization objective to attract customers; for each MT, the usage utility of the MS after completion can be expressed as equation (11), i.e., the usage utility of the selected MS:
Figure FDA0003816047030000042
lower task latency is an important indicator of meeting customer demand, whose value is related to the outcome of the service lease configuration;
for each MT, the overdue amount is a time interval exceeding the contracted period, and can be represented by (12):
Figure FDA0003816047030000043
(b) Decision vector y(τ)By aggregation of decision task queues
Figure FDA0003816047030000044
A medium-MT-related decision variable (14) component, indicating the processing start time of the MT (j, s):
Figure FDA0003816047030000045
Figure FDA0003816047030000046
wherein, the first and the second end of the pipe are connected with each other,
Figure FDA0003816047030000047
a decision variable indicating that the MT (j, s) selects for processing on the MS (i); as such, for a set of decision task queues
Figure FDA0003816047030000048
The optimization objective for the manufacturing service rental configuration can be expressed as (15):
Figure FDA0003816047030000049
establishing a constraint planning model, calling a mature solver to solve, and solving the solution to obtain
Figure FDA00038160470300000410
Figure FDA00038160470300000411
And attributes of MTRequired capacity
Figure FDA0003816047030000051
I.e. the rental configuration for the service.
CN201911030698.0A 2019-10-28 2019-10-28 Method for carrying out prospective cloud manufacturing service lease configuration based on demand prediction Active CN110866591B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201911030698.0A CN110866591B (en) 2019-10-28 2019-10-28 Method for carrying out prospective cloud manufacturing service lease configuration based on demand prediction

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201911030698.0A CN110866591B (en) 2019-10-28 2019-10-28 Method for carrying out prospective cloud manufacturing service lease configuration based on demand prediction

Publications (2)

Publication Number Publication Date
CN110866591A CN110866591A (en) 2020-03-06
CN110866591B true CN110866591B (en) 2022-11-01

Family

ID=69654680

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201911030698.0A Active CN110866591B (en) 2019-10-28 2019-10-28 Method for carrying out prospective cloud manufacturing service lease configuration based on demand prediction

Country Status (1)

Country Link
CN (1) CN110866591B (en)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111812982B (en) * 2020-07-13 2022-08-26 中国电子科技集团公司信息科学研究院 Cloud manufacturing resource dynamic sharing and intelligent allocation method for intelligent manufacturing

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2013102932A2 (en) * 2011-12-23 2013-07-11 Mzaya Private Limited System and method facilitating forecasting, optimization and visualization of energy data for an industry
CN106572187A (en) * 2016-11-11 2017-04-19 三维通信股份有限公司 Cloud manufacturing service scheduling method and system
CN106612323A (en) * 2016-08-08 2017-05-03 四川用联信息技术有限公司 Service resource integration optimization algorithm in cloud manufacturing
CN107392402A (en) * 2017-09-11 2017-11-24 合肥工业大学 Production and transport coordinated dispatching method and system based on modified Tabu search algorithm
CN108053047A (en) * 2017-10-26 2018-05-18 北京航天智造科技发展有限公司 Cloud resources of production scheduling methods, devices and systems
CN109643247A (en) * 2016-07-20 2019-04-16 世界线公司 The multi-standard adaptive scheduling of mixing cloud infrastructure for market orientation
CN109753026A (en) * 2017-11-07 2019-05-14 智能云科信息科技有限公司 Procreative collaboration method and system, storage medium, cloud manufacturing platform and production management system

Family Cites Families (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7650293B2 (en) * 2004-04-27 2010-01-19 Verint Americas, Inc. System and method for workforce requirements management
CN106611286A (en) * 2016-08-11 2017-05-03 四川用联信息技术有限公司 Dynamic allocation algorithm of service resources in cloud manufacturing
CN106503869A (en) * 2016-11-14 2017-03-15 东南大学 A kind of public bicycles dynamic dispatching method that is predicted based on website short-term needs
CN106960252A (en) * 2017-03-08 2017-07-18 深圳市景程信息科技有限公司 Methods of electric load forecasting based on long Memory Neural Networks in short-term
WO2019041000A1 (en) * 2017-09-01 2019-03-07 Go People Pty Ltd An intelligent demand predictive pre-emptive pre-sorting e-commerce order fulfilment, sorting and dispatch system for dispatch routing optimisation
CN109426920A (en) * 2018-01-19 2019-03-05 武汉十傅科技有限公司 A kind of enterprise's production planning optimization method considering prediction order and practical order
CN110059942B (en) * 2019-04-02 2022-08-30 南京邮电大学 Cloud manufacturing resource service optimization scheduling method based on fuzzy multi-objective optimization

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2013102932A2 (en) * 2011-12-23 2013-07-11 Mzaya Private Limited System and method facilitating forecasting, optimization and visualization of energy data for an industry
CN109643247A (en) * 2016-07-20 2019-04-16 世界线公司 The multi-standard adaptive scheduling of mixing cloud infrastructure for market orientation
CN106612323A (en) * 2016-08-08 2017-05-03 四川用联信息技术有限公司 Service resource integration optimization algorithm in cloud manufacturing
CN106572187A (en) * 2016-11-11 2017-04-19 三维通信股份有限公司 Cloud manufacturing service scheduling method and system
CN107392402A (en) * 2017-09-11 2017-11-24 合肥工业大学 Production and transport coordinated dispatching method and system based on modified Tabu search algorithm
CN108053047A (en) * 2017-10-26 2018-05-18 北京航天智造科技发展有限公司 Cloud resources of production scheduling methods, devices and systems
CN109753026A (en) * 2017-11-07 2019-05-14 智能云科信息科技有限公司 Procreative collaboration method and system, storage medium, cloud manufacturing platform and production management system

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
《Scheduling optimization of cloud resource supply chain through multi-objective particle swarm optimization》;H.Jiang et al.;《International Journal of Simulation Modelling》;20190131;全文 *
《云制造环境下的制造资源优化配置方法研究》;苏凯凯;《中国优秀博硕士学位论文全文数据库(博士)信息科技辑》;20170915(第9期);全文 *
《面向物流领域的云服务组合与调度优化方法》;初佃辉;《中国优秀博硕士学位论文全文数据库(博士)信息科技辑》;20160315(第3期);全文 *

Also Published As

Publication number Publication date
CN110866591A (en) 2020-03-06

Similar Documents

Publication Publication Date Title
CN110737529B (en) Short-time multi-variable-size data job cluster scheduling adaptive configuration method
Toka et al. Machine learning-based scaling management for kubernetes edge clusters
CN108009023B (en) Task scheduling method based on BP neural network time prediction in hybrid cloud
Adnan et al. Energy efficient geographical load balancing via dynamic deferral of workload
Jiang et al. Optimal cloud resource auto-scaling for web applications
Huang et al. Resource prediction based on double exponential smoothing in cloud computing
Salot A survey of various scheduling algorithm in cloud computing environment
CN101692208B (en) Task scheduling method for processing real-time traffic information
CN108965024A (en) A kind of virtual network function dispatching method of the 5G network slice based on prediction
CN108170531B (en) Cloud data center request flow scheduling method based on deep belief network
CN109740870B (en) Resource dynamic scheduling method for Web application in cloud computing environment
Zhao et al. Exploring fine-grained resource rental planning in cloud computing
Fard et al. Resource allocation mechanisms in cloud computing: a systematic literature review
CN109254836A (en) Time limit committed cost Optimization Scheduling towards the preferential dependence task of cloud computing system
Qian et al. A workflow-aided Internet of things paradigm with intelligent edge computing
CN109710372A (en) A kind of computation-intensive cloud workflow schedule method based on cat owl searching algorithm
Ye et al. SHWS: Stochastic hybrid workflows dynamic scheduling in cloud container services
CN109558991A (en) Based on cargo path amount recommended method, device, equipment and the storage medium from the machine of dealer
CN110866591B (en) Method for carrying out prospective cloud manufacturing service lease configuration based on demand prediction
Mazidi et al. An autonomic decision tree‐based and deadline‐constraint resource provisioning in cloud applications
CN115202847A (en) Task scheduling method and device
Dogani et al. K-agrued: a container autoscaling technique for cloud-based web applications in kubernetes using attention-based gru encoder-decoder
Nawrocki et al. Data-driven adaptive prediction of cloud resource usage
Mampage et al. Deep reinforcement learning for application scheduling in resource-constrained, multi-tenant serverless computing environments
Li et al. A QoS-based scheduling algorithm for instance-intensive workflows in cloud environment

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