CN113743728B - Cluster detection equalization scheduling method considering state transition - Google Patents

Cluster detection equalization scheduling method considering state transition Download PDF

Info

Publication number
CN113743728B
CN113743728B CN202110894537.7A CN202110894537A CN113743728B CN 113743728 B CN113743728 B CN 113743728B CN 202110894537 A CN202110894537 A CN 202110894537A CN 113743728 B CN113743728 B CN 113743728B
Authority
CN
China
Prior art keywords
detection
task
scheduling
state
time
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
CN202110894537.7A
Other languages
Chinese (zh)
Other versions
CN113743728A (en
Inventor
程颖
宋心怡
陶飞
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.)
Beihang University
Original Assignee
Beihang University
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 Beihang University filed Critical Beihang University
Priority to CN202110894537.7A priority Critical patent/CN113743728B/en
Publication of CN113743728A publication Critical patent/CN113743728A/en
Application granted granted Critical
Publication of CN113743728B publication Critical patent/CN113743728B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0631—Resource planning, allocation, distributing or scheduling for enterprises or organisations
    • G06Q10/06312—Adjustment or analysis of established resource schedule, e.g. resource or task levelling, or dynamic rescheduling
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Human Resources & Organizations (AREA)
  • Strategic Management (AREA)
  • Theoretical Computer Science (AREA)
  • Economics (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Software Systems (AREA)
  • Marketing (AREA)
  • Tourism & Hospitality (AREA)
  • Game Theory and Decision Science (AREA)
  • General Business, Economics & Management (AREA)
  • Operations Research (AREA)
  • Development Economics (AREA)
  • Quality & Reliability (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Medical Informatics (AREA)
  • Educational Administration (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention relates to a cluster detection equalization scheduling method considering state transition, which comprises the following steps: step 1, constructing a cluster aircraft engine fault detection problem description model; step 2, predicting the cluster arrival condition in the next scheduling period, namely the corresponding aero-engine fault detection task demand condition; step 3, evaluating the system state transition trend by combining the current detection task demand and the prediction condition of the system as well as the detection resources and the load state of the detection resources; step 4, configuring a balanced scheduling target facing different state transition trends, and generating a cluster aircraft engine fault detection balanced scheduling optimization problem model; and 5, outputting a fault detection and equalization scheduling scheme of the cluster aircraft engine. The invention realizes the homeopathic configuration of the system state equalization scheduling target by predicting the state transition trend of the cluster aircraft engine fault detection system, and can reduce the task delay rate and the system emergency response capability.

Description

Cluster detection equalization scheduling method considering state transition
Technical Field
The invention belongs to the technical field of intelligent scheduling for fault detection tasks of aircraft engine clusters, and particularly relates to a cluster detection equalization scheduling method considering state transition.
Background
Under the promotion of a new generation of detection technology and artificial intelligence technology, the fault detection of the aircraft engine is developed from the original artificial hole detection mode to an intelligent and mechanical fault detection mode taking aircraft engine fault detection automation equipment as a main factor. The fault detection of the aeroengine fault detection equipment resources is usually carried out in a combined and cooperative mode, visual inspection is carried out on parts such as turbine blades, combustion chambers and the like of the aeroengine, and when the aircraft fleet dynamically arrives to be detected, centralized and unified scheduling needs to be carried out on the detection resources.
The traditional scheduling method is mainly performed based on a single scheduling environment, however, under the conditions of uncertain cluster arrival time, uncertain cluster engine fault detection task arrival frequency, uncertain cluster fault scale and complex and changeable fault detection system comprehensive state, the traditional scheduling method cannot adapt to changeable scheduling system state scenes. The cluster fault detection system can not realize the long-term equalization target of the system state, so that the system has weak emergency capacity and low on-time task completion rate, and the traditional scheduling means is not suitable for the fault detection scene of the aircraft engine with uncertain conditions.
Disclosure of Invention
The technical problem to be solved by the invention is as follows: the cluster detection equalization scheduling method considering the state transition is based on prediction of a fault detection task of an aero-engine, evaluates the state transition trend of an aero-transmitter fault detection system, and configures different scheduling targets according to different system state transition conditions, so that long-term equalization of the system state is achieved, and the capability of the system for dealing with emergency situations and the on-time completion rate of the task are improved. The method comprises 5 steps of constructing a cluster detection problem model, predicting cluster detection task requirements, evaluating system state migration trend considering predicted requirements, configuring balanced scheduling targets facing different migration trends, and generating/updating a system state balanced scheduling scheme. The method can be applied to the dynamic scheduling of the detection resources of the aeroengine, realize the long-term equalization of the state of the cluster detection system, and improve the punctual completion rate of detection tasks and the emergency response capability of the detection system.
The technical problem to be solved by the invention is realized by adopting the following technical scheme: a cluster detection equalization scheduling method considering state transition comprises the following steps:
step 1: constructing detection resources and a task model, and expressing the number of K types of detection tasks in a scene as pkExtracting the attribute characteristics of the system detection task arriving within a period of time, fi kRepresenting the ith k-type task, the attribute characteristics including the detection of the task arrival time T _ releasek,iDetecting task deadline T _ duk,iDetecting task execution time T _ executek,iAnd the number of detection resources O corresponding to the taskk,i(ii) a For M types of detection resources in a scene, the number of M types of detection resources is denoted as qm，
Figure BDA0003197314710000021
Indicating a jth m-type detection resource;
and 2, step: predicting detection task information arriving in the next scheduling period based on historical detection task flow information through machine learning, wherein the prediction content comprises detection task related attributes;
and 3, step 3: comprehensively evaluating the system state according to the current task and resource state of the system, evaluating the system state at the next rescheduling moment based on the task prediction information obtained in the step (2), and evaluating the system migration condition through a set state threshold;
and 4, step 4: considering different system state transition conditions, selecting a combined configuration scheduling objective function by using a scheduling objective configuration method facing to detection system state equalization according to the system state transition conditions and taking the condition that the detection system state tends to be equalized as an objective, and constructing a fault detection task scheduling model by considering predictive scheduling or delay scheduling;
and 5: solving the problem by using an intelligent optimization algorithm based on the fault detection task scheduling model configured in the step 4; and finally, executing according to the solution scheme until the next scheduling moment triggers rescheduling, and repeating the processes of the steps 2-5 until all the detection tasks are executed.
Further, different types of detection resources and detection tasks may have different importance levels, that is, in the calculation of the urgency level and the scarcity level of the detection resources, different types of detection resources or detection task states may have different weight coefficients, which can be further discussed in a specific application scenario, and the method allocates an average weight coefficient.
Furthermore, for the division of the discrimination threshold of the scheduling scene, adjustment needs to be performed according to the actual application condition, and the method cannot provide a uniform standard. The task urgency degree value and the resource urgency degree value are both larger and more urgent, and the sum of the task urgency degree value and the resource urgency degree value can comprehensively represent the states of the tasks and the resources, so the method comprehensively represents the system urgency degree in the form of the sum.
Further, in the face of different system migration trends, the selection basis of the scheduling target of the method is as follows: when the system is relaxed, the detection resource is reserved, and when the system is urgent, the task is encouraged to be executed in a delayed way. The method can realize the equalization of the system state and the equalization of the supply level of the system resources to a certain extent, thereby improving the emergency response capability of the system.
Compared with the prior art, the invention has the advantages that:
(1) The comprehensive evaluation of the state of the detection system is considered, the state transition of the detection system is subjected to prediction evaluation on the basis that the machine learns the arrival information of the prediction task, and guidance is provided for selection of a scheduling target and generation of a scheduling scheme;
(2) In the face of different system state migration trends, the invention provides a scheduling target selection method facing system state equalization, different scheduling targets are configured and advanced and delayed scheduling is considered for optimization aiming at different scheduling situations, the emergency response capability of the system and the on-time completion rate of tasks can be effectively improved, and the scheduling method can also be applied to production scheduling systems with similar requirements.
Drawings
FIG. 1 is a flow chart of a cluster detection equalization scheduling method in accordance with the present invention that considers state transitions;
fig. 2 is a detailed flow chart of the present invention.
Detailed Description
The present invention is described in further detail below with reference to the attached drawings.
The invention discloses a cluster detection balanced scheduling method considering state transition, which comprises 5 steps of cluster detection problem model construction, cluster detection task demand prediction, system state transition trend evaluation considering prediction demand, balanced scheduling target configuration facing different transition trends and system state balanced scheduling scheme generation/update. The method can effectively predict the system state transition trend of the aircraft engine detection system, and can configure a proper scheduling target according to the state of the fault detection system so as to realize long-term equalization of the load state of the fault detection system, improve the task completion rate and the emergency response capability of the system, and ensure timely and efficient execution of the detection task.
The flow chart of the cluster detection equalization scheduling method considering state transition of the present invention is shown in fig. 1, the specific detailed flow chart is shown in fig. 2, and the specific implementation manner is as follows:
a cluster detection equalization scheduling method considering state transition comprises the following steps:
step 1: constructing detection resources and a task model, and expressing the number of K types of detection tasks in a scene as pkExtracting k-type detection task attribute features of the arrival system in a period of time, fi kRepresenting the ith k-type task, the attribute characteristics including the detection of the task arrival time T _ releasek,iDetecting task deadline T _ duk,iDetecting task execution time T _ executek,iAnd the number of detection resources O corresponding to the taskk,i(ii) a For M types of detection resources in a scene, the number of M types of detection resources is denoted as qm，
Figure BDA0003197314710000031
Indicating a jth m-type detection resource;
step 2: predicting detection tasks arriving in the next scheduling period based on historical detection task flow information through machine learning, wherein the prediction content comprises detection task related attributes;
and step 3: comprehensively evaluating the system state according to the task and resource state of the system at the moment, evaluating the system state at the next rescheduling moment based on the task prediction information obtained in the step 2, and evaluating the system migration condition through a set state threshold value;
and 4, step 4: considering the state transition conditions of different systems, selecting a combined configuration scheduling objective function by using a scheduling objective configuration method facing the state equalization of the detection system and aiming at achieving the condition that the state of the detection system tends to be equalized according to the state transition conditions of the system, and constructing a fault detection task scheduling model by considering predictive scheduling or delayed scheduling;
and 5: solving the problem by using an intelligent optimization algorithm based on the fault detection task scheduling model configured in the step 4; and finally, executing according to the solution scheme until the next scheduling moment triggers rescheduling, and repeating the processes of the steps 2-5 until all the detection tasks are executed.
Specifically, the first step: referring to fig. 1, a fleet detection problem model is constructed, and the specific implementation manner is as follows:
the step 1 specifically comprises:
1.1 during each dispatch cycle, a set of aircraft engine fault detection tasks is submitted to the dispatch system. For the detection tasks of K fault types in the scene, the number of the K types of tasks is represented as pkExtracting k-type detection task attribute features of the arrival system in a period of time, fi kRepresenting the ith k-type task, which can be represented by a quadruple, fi k＝{T_releasek,i,T_duek,i,T_execute k,i,Ok,iH, wherein T _ releasek,iFor task arrival time, T _ diek,iTo detect task deadline, T _ executek,iIndicating the duration of execution of the detection task, set
Figure BDA0003197314710000041
Represent a task fi kThe number of M detection resources required; for M types of detection resources in a scene, the number of M types of detection resources is denoted as qm，
Figure BDA0003197314710000042
Represents the jth m-type detection resource;
1.2 scheduling scheme for detection tasks, T _ startk,iIndicating a detection task fi kTime to begin execution, T _ endk,iIndicating the completion time of the detection task, Wk,iRepresentation detection task fi kDuration of queue wait to begin execution, Lk,iIndicating a detection task fi kA time delay;
1.3 is directed toTask execution process, method of setting parameterstAk,iUsed for judging whether the task reaches the system, if sotAk,i=1, representing a detection task fi kThe system is reached at the time t, otherwise, the system is not reached; setting parameterstXm,jFor indicating detection resources
Figure BDA0003197314710000043
Whether it is occupied or not, if it is occupiedtXm,j=1, otherwise thentXm,j=0,; setting parameter matrix
Figure BDA0003197314710000044
Used for representing the corresponding relation between the detection resource and the detection task if
Figure BDA0003197314710000045
Then the jth resource
Figure BDA0003197314710000046
Is being the ith task fi kService, otherwise, the service is not in the service; setting parameterstZk,iFor judging task fi kWhether it is executing, if it is executingtZk,i=1, otherwise thentZk,i＝0。
Step two: referring to fig. 1, 2, a cluster detection task demand prediction specifically includes the following steps:
the step predicts the cluster detection task information arriving in the next scheduling period based on the historical task information of the cluster detection system by means of machine learning prediction and the like, and the predicted task is recorded as
Figure BDA0003197314710000047
The total number of the predicted tasks is marked as P', wherein
Figure BDA0003197314710000051
Wherein k isi' to predict the task type of the detection task, T _ Releasei' to predict the arrival time of the detection task, T _ diei' to predict the deadline of the detection task, T _ executei' is to predict the execution time of the detection task, Oi' contains the required number of prediction tasks for each detection resource. Number of k-type detection tasks is pk' means.
Step three: referring to fig. 1, 3, considering the system state transition trend evaluation of the prediction demand, the specific implementation is as follows:
3.1. the method evaluates the urgency degree of the detection task at each time of scheduling execution. Equation (1) shows time k0The method for calculating the urgency degree of the type detection tasks considers the states of the different types of detection tasks from two aspects: on one hand, the time urgency of the type of detection task being executed in the detection system is considered, as shown in the first part of equation (1); on the other hand, the proportion of the type of tasks of the reached and unexecuted detection system among all the reached and unexecuted tasks is considered. Beta is a1,β2Are the weight coefficients. Equation (2) represents time k at t normalized by the range0Type of detection task urgency, andtF_statek∈[0,1]wherein the most urgent is 1 and the least urgent is 0 for each type of task.
Figure BDA0003197314710000052
Figure BDA0003197314710000053
The method also evaluates the detected resource shortage degree at each time when scheduling is executed. Expression (3) denotes time m0The method for calculating the type detection resource shortage degree considers the different types of detection resource shortage degrees from two aspects: on one hand, the occupied occupation ratio of the type of resources in the total resources is considered; on the other hand, consider the ratio of the time that the type of resource being occupied will also be occupied to the total occupied time. Alpha is alpha1,α2Are the weight coefficients. Equation (4) represents time m after t normalized by the range0Type of detecting resource scarcity, andtR_statm∈[0,1]wherein, the most scarce resource of each type is 1, and the least scarce resource of each type is 0.
Figure BDA0003197314710000054
Figure BDA0003197314710000061
And (5) evaluating the state of the detection system by integrating the urgency degree of each type of detection task and the resource shortage degree.tThe larger the value of S _ stat is, the more urgent the characterization detection system state is.
Figure BDA0003197314710000062
3.2. And (4) in combination with the detection task information predicted in the step (3), the method carries out prediction evaluation on the detection system state at the next rescheduling moment, namely the T + T moment, and the evaluation mode is similar to the step (3.1). Equation (6) is a calculation method for detecting the degree of task urgency in the system in consideration of the prediction information, and equation (7) is the degree of task urgency after normalization.
Figure BDA0003197314710000063
Figure BDA0003197314710000064
At this time, considering that the system detection resource shortage degree is not influenced by the prediction detection task information, the calculation modes can be obtained through deduction of the scheduling scheme at the time t, and are still represented by the formulas (3) and (4) and are recorded ast+TR_statem'. Thus, the (T + T) time detection system is evaluated in a comprehensive manner asFormula (8).
Figure BDA0003197314710000065
3.3. In the actual detection process of the aeroengine, a state judgment threshold value delta s is set according to expert experience knowledge, if the system state is smaller than delta s, the system is judged to be loose, and if the system state is larger than or equal to delta s, the system is judged to be urgent. By predictiont+TAfter S _ state', the system state transition trend can be evaluated, and the detected system transition can be evaluated as "slack-slack", "slack-urgent", "urgent-urgent", or "urgent-slack".
Step four: referring to fig. 1, 4, a configuration of balanced scheduling targets for different migration trends is specifically implemented as follows:
due to the fact that the system state equalization target is oriented, when the system state is relaxed, the system considers the reservation of detection resources, and when the system state is urgent, the system considers the reduction of the task delay rate. The selection of the scheduling target is shown in the following table:
Figure BDA0003197314710000071
aiming at the initial scheduling of the system, the task waiting time and the task delay time are reduced as scheduling targets, and the specific implementation mode is shown as formulas (9) and (10):
Figure BDA0003197314710000072
Figure BDA0003197314710000073
for the condition of 'relaxation-relaxation' of the detection system, at this time, the detection resources of the detection system are sufficient, the executable time of the detection task is sufficient, the detection system preferentially considers the improvement of the system response capability, namely, the scheduling target is set to minimize the urgency degree of the detection task and minimize the waiting time of the detection task, and the specific implementation method is shown in the formulas (9) and (11). Equation (11) indicates that the scheduling objective is to minimize the resource scarcity:
Figure BDA0003197314710000074
aiming at the 'loose-urgent' condition of the detection system, the detection resources of the detection system are still sufficient at the moment, the detection tasks of the detection system are obviously increased in the next period of time, the system needs to limit the resource occupancy rate and consider the prediction information, namely, the tasks in the next scheduling period are scheduled in advance through the detection task demand prediction. The specific implementation method is shown as a formula (11) and a formula (12). Equation (12) represents the minimization of the detection task urgency:
Figure BDA0003197314710000075
aiming at the situation of 'urgent-urgent' of the detection system, the detection resources of the detection system are in short supply, and the detection resources have no capability to be reserved for the detection tasks arriving at the next period of time. In consideration of improving task processing efficiency, setting a scheduling target to detect task completion delay time and minimize flow time, specifically implementing methods as shown in formulas (10) and (13), where formula (13) represents a scheduling target with minimized maximum flow time:
Figure BDA0003197314710000076
aiming at the 'urgent-loose' condition of the cluster detection system, the detection system detects the shortage of resources and has large detection task load capacity, so that the waiting time of the detection task with abundant time can be prolonged to a certain extent, and the detection task can be executed uniformly in the next scheduling period. The invention introduces a task delay scheduling mechanism with punishment, namely, under the condition of allowing the task deadline, delaying some detection tasks to the next period as much as possible and then scheduling, and setting the scheduling targetAnd (3) minimizing the delay time to sacrifice the response capability of the detection system, and exchanging for the reduction of the delay rate of the detection task, wherein the specific implementation forms are shown as formulas (10), (12), (14) and (15). Equation (14) is a calculation of the time Cost, where-c1·WiIs latency cost compensation, i.e., encouragement to delay the performance of the detection task. c. C1,c2∈[0,1]Are the weight coefficients.
Costk,i＝-c1·Wk,i+c2·Lk,i (14)
Equation (15) represents that the scheduling objective is time Cost minimization:
Figure BDA0003197314710000081
equation (16) represents the constraint on the total number of resources.
Figure BDA0003197314710000082
Step five: referring to fig. 1, 5, a system state equalization scheduling scheme is generated/updated, and the specific implementation manner is as follows:
based on the scheduling problem model, the problems are solved by adopting an intelligent solving algorithm, such as a particle swarm algorithm, a genetic algorithm, an evolutionary algorithm and the like, a scheduling rule and the like, the solving is carried out until the next rescheduling moment according to a solving scheme, and the flow of the second step to the fifth step is repeated until all tasks are executed.
In summary, the present invention discloses a cluster detection balancing scheduling method considering state migration, which includes 5 steps of cluster detection problem model construction, cluster detection task demand prediction, system state migration evaluation considering prediction demand, balancing scheduling target configuration facing different migration trends, and system load state balancing scheduling scheme generation/update. The method is applied to the detection scene of the aeroengine, is oriented to the load state balancing target of the detection system, and can effectively improve the task completion rate of the system for a long time and the emergency capacity of the system.
Those skilled in the art will appreciate that the invention may be practiced without these specific details.
The foregoing is only a preferred embodiment of the present invention, and it should be noted that, for those skilled in the art, various modifications and decorations can be made without departing from the principle of the present invention, and these modifications and decorations should also be regarded as the protection scope of the present invention.

Claims (5)

1. A cluster detection equalization scheduling method considering state transition is characterized by comprising the following steps:
step 1: constructing detection resources and a task model, and expressing the number of K types of detection tasks in a scene as pkExtracting attribute features of detection tasks arriving at the system over a period of time, wherein fi kRepresenting the ith k-type task, the attribute characteristics including detecting the arrival time T _ release of the taskk,iDetecting task deadline T _ duk,iDetecting task execution time T _ executek,iAnd corresponding number of detection resources Ok,i(ii) a For M types of detection resources in a scene, the number of M types of detection resources is denoted as qm，
Figure FDA0003834161830000011
Represents the jth m-type detection resource;
step 2: predicting detection tasks arriving in the next scheduling period based on historical detection task flow information through machine learning, wherein the prediction content comprises detection task related attributes;
and step 3: comprehensively evaluating the system state according to the task and resource state of the system at the moment, evaluating the system state at the next rescheduling moment based on the task prediction information obtained in the step 2, and evaluating the system migration condition through a set state threshold value; the step 3 specifically comprises:
3.1. at the time of rescheduling, toThe urgency degree of the detection task at this time is evaluated, and formula (1) represents k at this time t0Method for calculating the urgency of a type of task, beta1,β2Is a weight coefficient; equation (2) represents normalized k0Type of detection task urgency, andtF_statek∈[0,1]wherein each type of task is 1 most urgent and 0 least urgent;
Figure FDA0003834161830000012
Figure FDA0003834161830000013
wherein,tT_leftk,ias task fi kThe remaining execution time of; setting m at t considering the dynamic characteristics of the detection resource0The resource shortage status of type detection istr_statem0Formula (4) represents normalized k as shown in formula (3) and formula (4)0The degree of resource scarcity of the type is detected,tR_statem∈[0,1]wherein, the most scarce resource of each type of resource is 1, and the most abundant resource is 0;
Figure FDA0003834161830000014
Figure FDA0003834161830000015
wherein, T _ executek,iRepresenting a task fi kAn execution duration; alpha is alpha1,α2Is a weight coefficient; setting parameters for task execution processtAk,iUsed for judging whether the task reaches the system, if sotAk,i=1, representing a detection task fi kHas arrived at the system at time t, otherwiseThen it has not arrived; setting parameterstXm,jFor indicating detection resources
Figure FDA0003834161830000021
Whether or not it is occupied, and if sotXm,j=1, otherwise thentXm,j=0,; setting parameter matrix
Figure FDA0003834161830000022
Used for representing the corresponding relation between the detection resource and the detection task if
Figure FDA0003834161830000023
Then the jth resource
Figure FDA0003834161830000024
Is being the ith task fi kService, otherwise, the service is not in the service; setting parameterstZk,iFor judging task fi kWhether or not it is executing, iftZk,iIf not, the task is executed, otherwise, the task is executedtZk,i＝0；
The formula (5) evaluates the state of the detection system by considering the urgency degree of each type of detection task and the shortage degree of detection resources,tthe larger the S _ state value is, the more urgent the representation detection system state is;
Figure FDA0003834161830000025
3.2. according to the detection task information of the next scheduling period T arriving at the system predicted in the step 2, the detection tasks and the detection resource shortage state are evaluated, and then the comprehensive state of the detection system at the moment (T + T) is evaluated, firstly, the detection resources occupied by the detection system at the moment (T + T) and the detection tasks being executed are calculated according to the prediction information, and the comprehensive state of the detection system is set to bet+TS_state′；
Equation (6) is a calculation method of the system detection task urgency degree in consideration of the prediction information, and equation (7) is the task urgency degree after the standardization;
Figure FDA0003834161830000026
Figure FDA0003834161830000027
at this time, considering that the system detection resource shortage degree is not influenced by the prediction detection task information, the system detection resource shortage degree can be obtained after the deduction of the scheduling scheme at the time t and is marked ast+TR_statem', whereby the overall state evaluation of the detection system at time (T + T) is shown in equation (8):
Figure FDA0003834161830000028
in the actual cluster fault detection process, a system state threshold value deltas is set according to the experience of experts and is used for defining the urgent state and the loose state of the system state, the slack state is set when the system state is lower than the threshold value, and the urgent state is set when the system state is larger than or equal to the threshold value; judging the system state to be loose or urgent through a set system state threshold value, further deducing the system state transition trend, and expressing the system state at the moment-the next scheduling moment state of the system as follows: "relaxed-relaxed", "relaxed-urgent", "urgent-relaxed";
and 4, step 4: considering the state transition conditions of different systems, selecting a configuration scheduling objective function by utilizing a scheduling objective configuration method facing the state equalization of the detection system according to the state transition conditions of the system to achieve the aim of detecting the state of the system tending to the equalization, and constructing a fault detection task scheduling model by considering predictive scheduling or delayed scheduling;
and 5: solving the problem by using an intelligent optimization algorithm based on the fault detection task scheduling model configured in the step 4; and finally, executing according to the solution scheme until the next scheduling moment triggers rescheduling, and repeating the processes of the steps 2-5 until all the detection tasks are executed.
2. The cluster detection equalization scheduling method in consideration of state migration as claimed in claim 1, wherein:
the step 1 specifically comprises:
1.1 during each scheduling cycle, a set of aeroengine fault detection tasks is submitted to the scheduling system, the number of K types of tasks in the scene is represented as pkExtracting detection task attribute features of the arrival system within a period of time, fi kRepresenting the ith k-type task, which can be represented by a quadruple, fi k＝{T_releasek,i,T_duek,i,T_executek,i,Ok,iWherein T _ releasek,iFor task arrival time, T _ diek,iAs task fi kCutoff time, T _ executek,iRepresenting a task fi kDuration of execution, set
Figure FDA0003834161830000031
Represents a task fi kThe number of M types of detection resources required; for M types of detection resources in a scene, the number of M types of detection resources is denoted as qm，
Figure FDA0003834161830000032
Represents the jth m-type detection resource;
1.2 scheduling scheme for detection tasks, T _ startk,iRepresentation detection task fi kTime to begin execution, T _ endk,iIndicating a detection task fi kCompletion time, Wk,iRepresentation detection task fi kDuration of queue wait to begin execution, Lk,iIndicating a detection task fi kA delay time duration;
1.3 setting parameters for task executiontAk,iUsed for judging whether the task reaches the system, if sotAk,i=1, representing a detection task fi kThe system is reached at the time t, otherwise, the system is not reached; setting parameterstXm,jFor indicating detection resources
Figure FDA0003834161830000033
Whether it is occupied or not, if it is occupiedtXm,j=1, otherwise thentXm,j=0,; setting parameter matrix
Figure FDA0003834161830000034
Is used for representing the corresponding relation between the detection resource and the detection task if
Figure FDA0003834161830000035
Then the jth resource
Figure FDA0003834161830000036
Is being the ith task fi kService, otherwise, the service is not in the service; setting parameterstZk,iFor judging task fi kWhether it is executing, iftZk,iIf not, the task is executed, otherwise, the task is executedtZk,i＝0。
3. The cluster detection equalization scheduling method considering state migration according to claim 1, wherein: the step 2 specifically comprises:
training a machine learning model by using the historical task information of the cluster detection system, and predicting the task arrival time and the task type information in the next scheduling period T at each detection task rescheduling moment; the predicted task is recorded as
Figure FDA0003834161830000041
The total number of the predicted tasks is marked as P', wherein
Figure FDA0003834161830000042
The prediction content of the detection task comprises the type of the detection task, the arrival time of the detection task, the deadline of the detection task, the predicted execution time of the detection task and the predicted quantity of the required detection resources.
4. The cluster detection equalization scheduling method considering state migration according to claim 1, wherein: the step 4 mainly comprises:
according to the cluster detection system state transition trend evaluated in the previous step, a proper scheduling target is configured to enable the system state to achieve long-term balance, and the method specifically comprises the following steps:
aiming at the 'relaxation-relaxation' condition, the detection system has sufficient detection resources and sufficient detection task executable time, the detection system gives priority to improving the system response speed, and the scheduling target is set to reduce the task waiting time and reduce the resource shortage degree;
aiming at the 'loose-urgent' condition of the detection system, the detection resources of the detection system are sufficient at the moment, the load capacity of the detection tasks of the detection system is obviously increased in the next period of time, the prediction scheduling needs to be considered at the moment, namely, the tasks in the next scheduling period are scheduled in advance through the detection task demand prediction, and the scheduling target is set to reduce the resource shortage degree and the task urgent degree;
aiming at the situation of 'urgent-urgent' of the detection system, the detection resources of the detection system are unavailable and reserved for the detection task arriving at the next period of time; at the moment, the task processing efficiency is considered to be improved, and a scheduling target is set to minimize the task delay time and minimize the task maximum flow time;
aiming at the 'urgent-loose' condition of the detection system, the detection task needs to be uniformly executed in the next scheduling period, a task delay scheduling mechanism with punishment is introduced, namely, under the condition that the task deadline allows, the task is delayed to the next period for scheduling, the scheduling target is set to minimize the delay time, and the task urgency degree is reduced.
5. The cluster detection equalization scheduling method considering state migration according to claim 1, wherein: the step 5 mainly comprises:
and according to the scheduling target selected in the last step, selecting an intelligent optimization algorithm comprising a particle swarm algorithm, a genetic algorithm and a simulated annealing algorithm, solving the scheduling problem, fixing a scheduling period T, triggering one-time rescheduling every time one period T passes, simultaneously performing prediction evaluation on the state transition of the detection system, and updating and executing the scheduling scheme after the solution.
CN202110894537.7A 2021-08-05 2021-08-05 Cluster detection equalization scheduling method considering state transition Active CN113743728B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110894537.7A CN113743728B (en) 2021-08-05 2021-08-05 Cluster detection equalization scheduling method considering state transition

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110894537.7A CN113743728B (en) 2021-08-05 2021-08-05 Cluster detection equalization scheduling method considering state transition

Publications (2)

Publication Number Publication Date
CN113743728A CN113743728A (en) 2021-12-03
CN113743728B true CN113743728B (en) 2022-11-01

Family

ID=78730133

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110894537.7A Active CN113743728B (en) 2021-08-05 2021-08-05 Cluster detection equalization scheduling method considering state transition

Country Status (1)

Country Link
CN (1) CN113743728B (en)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114493100A (en) * 2021-12-21 2022-05-13 北京航空航天大学 Cluster detection situation identification and task generation method
CN114971162B (en) * 2022-04-12 2023-01-17 北京航空航天大学 State-adaptive dynamic scheduling method for cluster engine detection tasks

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104077393B (en) * 2014-07-01 2017-07-11 国家电网公司 A kind of optimal splitting fracture surface searching method based on semi-supervised spectral clustering
CN106936892A (en) * 2017-01-09 2017-07-07 北京邮电大学 A kind of self-organizing cloud multi-to-multi computation migration method and system
WO2018206994A1 (en) * 2017-05-11 2018-11-15 Bull Sas Method of managing resource providing in a computers cluster running jobs
CN108196958B (en) * 2017-12-29 2020-09-29 北京泽塔云科技股份有限公司 Resource scheduling and distributing method, computer system and super-fusion architecture system
CN112269632B (en) * 2020-09-25 2024-02-23 北京航空航天大学杭州创新研究院 Scheduling method and system for optimizing cloud data center

Also Published As

Publication number Publication date
CN113743728A (en) 2021-12-03

Similar Documents

Publication Publication Date Title
CN113743728B (en) Cluster detection equalization scheduling method considering state transition
CN109960585B (en) Resource scheduling method based on kubernets
CN110825520B (en) Cluster extremely-fast elastic telescoping method for realizing efficient resource utilization
CN104915407B (en) A kind of resource regulating method based under Hadoop multi-job environment
CN109324875B (en) Data center server power consumption management and optimization method based on reinforcement learning
CN110737529A (en) cluster scheduling adaptive configuration method for short-time multiple variable-size data jobs
CN114756358B (en) DAG task scheduling method, device, equipment and storage medium
CN110096349A (en) A kind of job scheduling method based on the prediction of clustered node load condition
US20220012089A1 (en) System for computational resource prediction and subsequent workload provisioning
CN108650131B (en) Processing system for multi-controller deployment in SDN network
CN109960573B (en) Cross-domain computing task scheduling method and system based on intelligent perception
CN113240287B (en) Manufacturing service self-adaptive scheduling method based on supply and demand relationship
CN106600446B (en) Production field event real-time processing system and method based on dynamic priority
CN112433807A (en) Airflow perception type virtual machine scheduling method oriented to data center global energy consumption optimization
KR20240105337A (en) Method and apparatus for resource allocation in cluster computing system
CN111782394A (en) Cluster service resource dynamic adjustment method based on response time perception
CN116627237B (en) Power management chip architecture system based on core chip
CN113094155B (en) Task scheduling method and device under Hadoop platform
CN109271295B (en) Abnormal operation prediction method in cloud cluster environment
CN116069473A (en) Deep reinforcement learning-based Yarn cluster workflow scheduling method
CN112379598A (en) Multi-gate-hole gate scheduling method based on dynamic programming algorithm
CN114971162B (en) State-adaptive dynamic scheduling method for cluster engine detection tasks
CN113010319A (en) Dynamic workflow scheduling optimization method based on hybrid heuristic rule and genetic algorithm
Putz et al. Simulation-Based Optimization Approach for Efficient Generation of Sequencing Heuristics for Solving the Stochastic Resource-Constrained Scheduling Problem
CN117271093A (en) Joint optimization scheduling method taking degree of freedom of edge scheduling as core

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