CN113158435A - Complex system simulation running time prediction method and device based on ensemble learning - Google Patents
Complex system simulation running time prediction method and device based on ensemble learning Download PDFInfo
- Publication number
- CN113158435A CN113158435A CN202110330626.9A CN202110330626A CN113158435A CN 113158435 A CN113158435 A CN 113158435A CN 202110330626 A CN202110330626 A CN 202110330626A CN 113158435 A CN113158435 A CN 113158435A
- Authority
- CN
- China
- Prior art keywords
- model
- candidate
- simulation
- parameters
- 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.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
-
- 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"
-
- 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
Abstract
The application relates to a complex system simulation running time prediction method and device based on ensemble learning, computer equipment and a storage medium. The method comprises the following steps: collecting operating parameters of simulation application operation to obtain mold-entering operating parameters; generating a candidate model combination and acquiring a preset meta-model to be trained; performing integrated training on each candidate model combination and the meta-model to be trained by using the mold-entering operation parameters to obtain candidate integrated models; and evaluating each candidate integration model, determining a simulation run-time prediction model, and predicting the run time of the simulation application by using the run-time prediction model. By adopting the method, efficient resource management can be realized.
Description
Technical Field
The application relates to the technical field of cloud simulation, in particular to a complex system simulation running time prediction method and device based on ensemble learning, computer equipment and a storage medium.
Background
Complex System simulation (CSM) is widely used for system evaluation and analysis in the fields of computer and telecommunication systems, biological networks, military training and war games. In CSM, a physical system is modeled as a plurality of logical processes (LPs, also referred to as simulation entities), with simulation occurring as a sequence of discrete events processed by LPs. However, complex system simulations tend to be composed of a large number of entities and there is a complex interaction between the entities, resulting in irregular workload and communication load and reducing the operating efficiency of the underlying infrastructure. In addition, to cope with huge workload changes, reduce simulation runtime, and meet new system requirements in terms of computational and memory resources, powerful computing infrastructure is required to perform correctly. Not all organizations have access to unlimited computing resources, so budget constraints are always an important consideration. At the same time, given a resource or budget, it is often desirable that an application be executed in the shortest possible time. Therefore, there is a need for a flexible infrastructure deployment mechanism that can allocate resources according to the computational requirements of an application.
At present, the cloud computing environment can realize cooperative management, demand sharing and flexible scheduling of resources such as computing/network/software/model and the like, and meets the requirements of complex system simulation on general computing power and efficient operation of simulation application, so that the cloud computing environment provides a good target environment for the challenges. To make efficient use of this environment, the simulation entities in a CSM are typically divided into groups, each of which is mapped to a node. In the running process of the simulation application, the running performance of the application is monitored and predicted in real time, so that real-time resource scheduling is carried out in a cloud computing environment, and the overall deployment cost is reduced.
However, the conventional resource scheduling process relies heavily on task run-time estimation to make decisions. In the execution process of the simulation application of the complex system, the demands of the simulation entity on computing resources are continuously changed due to different life cycles of the simulation entity, so that the running time of the simulation application in the cloud environment is difficult to accurately predict by the traditional method. Furthermore, the resources are allocated according to inaccurate running time, and if the resources allocated to the application are too small, it is difficult to support efficient running of the application. If the resources allocated to the application are too many, the communication overhead between different node entities is increased, and one of the resources is unnecessarily wasted. Therefore, how to accurately predict the execution time of the simulation application in the cloud environment, and realizing efficient resource management of the cloud simulation application is a challenging task.
Disclosure of Invention
In view of the foregoing, it is desirable to provide a method, an apparatus, a computer device and a storage medium for predicting simulation runtime of a complex system based on ensemble learning, which can achieve efficient resource management.
A method for ensemble learning based runtime prediction of a complex system simulation, the method comprising:
collecting operating parameters of simulation application operation to obtain mold-entering operating parameters;
generating a candidate model combination and acquiring a preset meta-model to be trained;
performing integrated training on each candidate model combination and the meta-model to be trained by using the mold-entering operation parameters to obtain candidate integrated models;
and evaluating each candidate integration model, determining a simulation run-time prediction model, and predicting the run time of the simulation application by using the run-time prediction model.
In one embodiment, the collecting the operation parameters of the simulation application operation to obtain the in-mode operation parameters includes:
monitoring and collecting the pre-operation parameters and the operation parameters of the simulation application by using the cloud computing nodes to obtain the operation parameters of the simulation application;
and selecting the characteristics of the operation parameters, and screening to obtain the mold-entering operation parameters.
In one embodiment, the selecting the characteristics of the operating parameters and screening to obtain the in-mold operating parameters includes:
respectively carrying out feature selection on each operation parameter by using a variance selection method to obtain the feature importance of each operation parameter;
and sorting the operation parameters according to the feature importance screening, and screening to obtain the in-mold operation parameters.
In one embodiment, the evaluating each of the candidate integration models to determine a simulation runtime prediction model includes:
each candidate integration model respectively utilizes the verification operation parameters to predict the operation time to obtain the evaluation prediction time;
determining the real running time corresponding to the verification running parameters, and calculating the average value of the real running times to obtain the average running time;
obtaining an evaluation value of each candidate integration model according to the evaluation prediction time, the real running time and the average running time;
and determining a simulation running time prediction model according to the evaluation value of each candidate integration model.
In one embodiment, the performing, by using the mold-entry operating parameter, integrated training on each candidate model combination and the meta-model to be trained to obtain a candidate integrated model includes:
respectively and independently training each model in the candidate model combination by taking the mold-entering operation parameter as input and the operation time corresponding to the mold-entering operation parameter as a target;
respectively predicting the running time by using each model in the trained candidate model combination;
fusing the operation time predicted by each model in the candidate model combination and the input operation parameters corresponding to the predicted operation time into an original training set to obtain a fused training set;
and training the meta-model to be trained according to the fusion training set to obtain a trained candidate integrated model.
In one embodiment, each model in the model set is independently trained using K-fold cross-validation.
In one embodiment, the generating a candidate model combination includes:
acquiring a configured model set;
and randomly selecting a preset number of models from the model set to form candidate model combinations.
An ensemble learning based complex system simulation runtime prediction apparatus, the apparatus comprising:
the data collection module is used for collecting the running parameters of the running of the simulation application to obtain the in-mold running parameters;
the model generation module is used for generating a candidate model combination and acquiring a preset meta-model to be trained;
the integrated training module is used for performing integrated training on each candidate model combination and the meta-model to be trained by using the in-mode operation parameters to obtain a candidate integrated model;
and the evaluation application module is used for evaluating each candidate integration model, determining a simulation running time prediction model and predicting the running time of the simulation application by using the running time prediction model.
A computer device comprising a memory storing a computer program and a processor implementing the steps of any of the above-described methods of ensemble learning based complex system simulation run-time prediction when the computer program is executed.
A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the steps of the method for runtime prediction of a complex system simulation based on ensemble learning of any of the above.
The complex system simulation running time prediction method and device based on ensemble learning, the computer equipment and the storage medium obtain the in-mode running parameters by collecting the running parameters of simulation application running; generating a candidate model combination and acquiring a preset meta-model to be trained; performing integrated training on each candidate model combination and the meta-model to be trained by using the in-mold operation parameters to obtain a candidate integrated model; and then evaluating each candidate integration model, determining a simulation run-time prediction model, and predicting the run time of the simulation application by using the run-time prediction model. According to the method, the model for predicting the simulation application running time in the cloud computing environment is built through the ensemble learning method to predict the cloud simulation application running time, so that efficient resource management is achieved for the cloud simulation application.
Drawings
FIG. 1 is a diagram of an application environment of a method for runtime prediction of simulation of a complex system based on ensemble learning in one embodiment;
FIG. 2 is a schematic flow chart illustrating a method for runtime prediction of simulation of a complex system based on ensemble learning in one embodiment;
FIG. 3 is a schematic representation of the ordering of feature importance of operational parameters in one embodiment;
FIG. 4 is a flow diagram illustrating ensemble learning prediction in one embodiment;
FIG. 5 is a graph of actual simulation run time versus predicted time in one embodiment;
FIG. 6 is a block diagram of an apparatus for runtime prediction of complex system simulation based on ensemble learning according to an embodiment;
FIG. 7 is a diagram illustrating an internal structure of a computer device according to an embodiment.
Detailed Description
In order to make the objects, technical solutions and advantages of the present application more apparent, the present application is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present application and are not intended to limit the present application.
The method for predicting the simulation running time of the complex system based on the ensemble learning can be applied to the application environment shown in FIG. 1. Wherein the terminal 102 communicates with the server 104 via a network. When the terminal 102 receives a prediction instruction issued by a user, the simulation run time prediction method can be implemented by the terminal 102 alone. Alternatively, the terminal 102 sends the prediction instruction to the server 104, and the server 104 implements the simulation runtime prediction method. Specifically, taking the server 104 as an example, the server 104 collects the running parameters of the simulation application running to obtain the in-mold running parameters; the server 104 generates a candidate model combination and acquires a preset meta-model to be trained; the server 104 performs integrated training on each candidate model combination and the meta-model to be trained by using the in-mold operation parameters to obtain a candidate integrated model; the server 104 evaluates each candidate integration model, determines a simulation run-time prediction model, and predicts the run-time of the simulation application using the run-time prediction model. And the server 104 stores the obtained simulation run time prediction model, and when a prediction instruction is received subsequently, the server 104 can directly call the simulation run time prediction model to predict the run time. The terminal 102 may be, but not limited to, various personal computers, notebook computers, smart phones, tablet computers, and portable wearable devices, and the server 104 may be implemented by an independent server or a server cluster formed by a plurality of servers.
In one embodiment, as shown in fig. 2, there is provided a method for predicting runtime of a complex system simulation based on ensemble learning, which is described by taking the method as an example applied to the server in fig. 1, and includes the following steps:
step S202, collecting real-time operation parameters of the simulation application program to obtain the mold-in operation parameters.
The operation parameters are real-time operation information of the simulation application program and comprise pre-operation parameters and operation parameters. The in-mold operating parameters are selected operating parameters for training the model, and may be all operating parameters or some operating parameters obtained by screening from the operating parameters.
Specifically, the server monitors and collects real-time running information of the simulation application program through the monitoring program to obtain running parameters. And then selecting the mold-entering operation parameters from the collected operation parameters according to the requirements.
And step S204, generating a candidate model combination and acquiring a preset meta-model to be trained.
The candidate model combination and the meta-model to be trained are models which are ready to be trained by using the in-mode operation parameters. The candidate model combination includes a plurality of different types of models. In this embodiment, the meta-model to be trained is preferably a random forest model.
Specifically, the server obtains a pre-configured model set, and screens a plurality of different sets of model combinations from the model set as candidate model combinations. And meanwhile, acquiring a pre-configured meta-model to be trained. In order to better predict the simulation application running time, when the model in the model set is selected to be the combination candidate model combination, the model needs to be automatically parametrized.
And S206, performing integrated training on each candidate model combination and the meta-model to be trained by using the mold-entering operation parameters to obtain a candidate integrated model.
The integrated training refers to an integrated learning process for combining a plurality of weak supervision models to obtain a better and more comprehensive strong supervision model.
Specifically, the server firstly trains each model in the candidate model combination by using the in-mold operation parameters, then integrally trains the meta-model to be trained based on the prediction capability of each trained model to obtain the candidate integrated models, and finally the number of the candidate integrated models is the same as the number of the generated candidate model combinations.
And S208, evaluating each candidate integration model, determining a simulation running time prediction model, and predicting the running time of the simulation application by using the running time prediction model.
Specifically, after a plurality of candidate integration models are obtained through integration training, the candidate integration model with the highest prediction capability is determined as a trained simulation runtime prediction model by evaluating the prediction capability of each candidate integration model. The server then uses the simulation run-time prediction model to make predictions of the simulation application run-time. That is, when the runtime of a certain simulation application needs to be predicted, the runtime parameters of the simulation application are input to the simulation runtime prediction model to obtain the predicted runtime.
According to the method for predicting the simulation running time of the complex system based on the ensemble learning, the model-entering running parameters are obtained by collecting the running parameters of the simulation application running; generating a candidate model combination and acquiring a preset meta-model to be trained; performing integrated training on each candidate model combination and the meta-model to be trained by using the in-mold operation parameters to obtain a candidate integrated model; and then evaluating each candidate integration model, determining a simulation run-time prediction model, and predicting the run time of the simulation application by using the run-time prediction model. According to the method, the model for predicting the simulation application running time in the cloud computing environment is built through the ensemble learning method to predict the cloud simulation application running time, so that efficient resource management is achieved for the cloud simulation application.
In one embodiment, step S202 includes: monitoring and collecting the pre-operation parameters and the operation parameters of the simulation application by using the cloud computing nodes to obtain the operation parameters of the simulation application; and selecting the characteristics of the operation parameters, and screening to obtain the mold-entering operation parameters.
Specifically, in order to accurately monitor and collect real-time operation information of a simulation application, a resource monitor is deployed on a cloud computing node, and an operation parameter is collected by using the resource monitor deployed on the cloud computing node. The resource monitor collects information about every 5 seconds and stores the information as feature data into a database after normalization processing. The present embodiment considers two sets of parameters to evaluate the performance of the application: pre-run parameters and run-time parameters, respectively. The parameters before running are determined before the simulation application is executed, and comprise the number of computing nodes for executing the simulation application, the number of simulation application entities, a look-ahead value, simulation time and the like. The runtime parameters reflect the performance difference of different parameter simulation applications under specific cloud resources, and are collected in the process of executing the simulation applications, including the CPU utilization, the memory utilization, the network throughput, the network latency, the file system utilization, and the like. And then, the server selects the characteristics of the collected operation parameters to obtain proper mold-entering operation parameters capable of improving the model capability.
In one embodiment, the selecting the characteristics of the operation parameters and screening to obtain the in-mold operation parameters includes: respectively selecting the characteristics of each operation parameter by using a variance selection method to obtain the characteristic importance of each operation parameter; and sorting the operation parameters according to the feature importance screening, and screening to obtain the in-mold operation parameters.
In particular, feature selection as a data preprocessing strategy has proven effective in data for various data mining and machine learning problems. The purpose of feature selection is to build simpler, more understandable models and improve the performance of the models. Feature extraction generally applies a mathematical method to map data from a high-dimensional feature space to a low-dimensional feature space, and converts original features that cannot be identified by a machine learning algorithm into new features that can be identified. However, because the data features in the simulation application log dataset are all represented in numerical form, no conversion is required. Therefore, the present embodiment directly applies the feature selection method to remove irrelevant features, and selects strongly relevant features and weakly relevant but non-redundant features, so as to minimize the occurrence of errors and build a more accurate prediction model. Therefore, in the present embodiment, it is preferable to perform feature selection on each operating parameter based on the variance selection method, so as to obtain the feature importance of each operating parameter. The features shown in FIG. 3 are then used as inputs to the predictive model, i.e., the in-mode operating parameters, according to a ranking of feature importance, as shown in FIG. 3. The description of the operating parameters ultimately selected for this embodiment as the in-mold operating parameters is shown in table 1 below.
Table 1: description of mold-in-operation parameters
In one embodiment, evaluating each candidate integration model to determine a simulation runtime prediction model comprises: each candidate integration model respectively utilizes the verification operation parameters to predict the operation time to obtain the evaluation prediction time; determining the real running time corresponding to the verification running parameters, and calculating the average value of each real running time to obtain the average running time; obtaining the evaluation value of each candidate integration model according to the evaluation prediction time, the real running time and the average running time; and determining a simulation running time prediction model according to the evaluation value of each candidate integration model.
The verification operation parameters are collected operation parameters used for estimating the prediction capability of the candidate integration model, and also comprise pre-operation parameters and operation parameters. Each set of operating parameters has a corresponding real operating time, i.e., an operating time actually corresponding to the set of operating parameters.
Specifically, after a plurality of candidate integration models are obtained through training, in order to select an optimal candidate integration model as a model for predicting the running time, the server respectively predicts the running time of each candidate integration model through verifying the running parameters, and thus the evaluation prediction time of each candidate integration model is obtained. And evaluating each candidate integration model according to the evaluation prediction time and the real running time. The server firstly calculates the average value of each real running time corresponding to each verification running parameter to obtain the average running time. Then, an evaluation value R of each candidate integration model is obtained according to the evaluation prediction time, the real running time and the average running time2The calculation formula is as follows:
wherein the content of the first and second substances,
and
respectively representing the real operation time and the estimated prediction time of the jth group of verified operation parameters,
the average run time is indicated. Finally, based on the evaluation value R2Selecting the best candidate integration model as the simulation run-time prediction model, e.g. determining the evaluation value R2The highest candidate integration model serves as a simulation runtime prediction model. The combination of candidate models corresponding to the best candidate integrated model determined by the evaluation includes ten models, and the ten models and the model use parameters are specifically as shown in table 2 below.
TABLE 2 machine learning model
In one embodiment, step S206 includes: respectively and independently training each model in the candidate model combination by taking the operation parameters of the mold entering as input and the operation time corresponding to the operation parameters of the mold entering as a target; respectively predicting the running time by using each model in the trained candidate model combination; fusing the operation time predicted by each model in the candidate model combination and the input operation parameters corresponding to the predicted operation time into an original training set to obtain a fused training set; and training the meta-model to be trained according to the fusion training set to obtain a trained candidate integrated model.
Specifically, as shown in fig. 4, a schematic diagram of an ensemble learning prediction process is provided, and referring to fig. 4, after completing collection and generation of a candidate model combination, a server first uses a model-entry operation parameter as an input of a model, and an operation time corresponding to the model-entry operation parameter is a target, and trains each model in the candidate model combination independently to complete training of each model in the candidate model combination. The present embodiment preferably trains each model in the model set independently using K-fold cross validation. And then, collecting a batch of new operation parameters as input by the server to predict the operation time of each model in the trained candidate model combination, so as to obtain the predicted operation time of each model. And the server adds the running time obtained by predicting each model and the running parameters input corresponding to the running time obtained by predicting, namely the collected new batch of running parameters into the original training set to obtain a new fusion training set. The original training set comprises the mold-entering operation parameters acquired by previous training and the operation time corresponding to the mold-entering operation parameters. The fusion training set comprises the sample data of the original training set, the predicted operation time of each model added at this time and the corresponding input operation parameters. And after the fusion training set is obtained, the server trains the meta-model to be trained by taking the operation parameters in the fusion training set as input and the corresponding operation time as a target to obtain a candidate integrated model. In the embodiment, the integrated model is trained in an integrated learning mode, so that the defect of single basic prediction can be improved, and the prediction capability is improved by utilizing the interaction between the basic models.
In addition, to evaluate the effectiveness of this integrated training in detail, we performed a number of experiments. The performance of each machine-learned regression model was first measured and the integrated model was then evaluated, with the experimental results shown in table 3 below.
TABLE 3 model Performance evaluation
| Model name | R2 | RMSE | MAE | ACC(％) |
| KNN | 0.817 | 97.79 | 62.13 | 75.31 |
| SVR | 0.942 | 53.28 | 34.83 | 81.59 |
| MLP | 0.948 | 53.03 | 38.23 | 83.35 |
| LR | 0.946 | 54.61 | 37.50 | 83.37 |
| DT | 0.957 | 48.67 | 36.43 | 84.45 |
| ETR | 0.962 | 45.71 | 32.03 | 86.37 |
| RF | 0.963 | 42.17 | 28.17 | 89.81 |
| XBG | 0.966 | 40.01 | 28.30 | 90.21 |
| SEP | 0.977 | 36.56 | 23.69 | 89.54 |
| SRPA | 0.972 | 37.7 | 22.04 | 93.72 |
The SRPA is a model obtained by an integration algorithm selected by model combination provided in this embodiment. The SEP is a model representing an integration algorithm without a model combination selection process. The calculation formulas of the Mean Absolute Error (MAE), the precision (Accuracy) and the Root Mean Square Error (RMSE) are as follows:
from the above, it can be seen that, with respect to a single model, each model has different performance in any evaluation index. More specifically, a single predictive model may be better than other predictive models in terms of error rate, but may have poorer accuracy or higher execution time. For example, the accuracy of the LR model is 83% higher than that of the SVR model, but the RMSE and MAE values are 54.61 and 37.50, respectively, which are also higher than the error rate of the SVR model. In addition, the present embodiment also tests the integrated model without the model selection process, and it can be found that the SRPA model has a higher prediction accuracy (93.72%) when the error rate (37.7/22.04) is equivalent to the SEP model.
Moreover, the present embodiment also runs the hold applications (hold is usually used as a representative benchmark test program for PDES performance evaluation) with two different simulation event numbers (50 and 100, respectively) in the cloud environment, and executes them in parallel using different cpu core numbers. As can be seen from fig. 5, the SRPA predicts both runtimes close to the true values with a maximum error of ± 30 seconds. Furthermore, when the number of the phosphor simulation events is 50, the application runs the shortest with 7 cores allocated. However, as the hold simulation event increases to 100, the resources required to apply the shortest runtime decrease to 6 cores.
In summary, the prediction result shows that the method provided by the embodiment can effectively predict the running time of the simulation application and select the optimal computing resource for the application. Compared with the existing machine learning algorithm, the algorithm provided by the method can improve the prediction accuracy by 3% -24% while keeping the lowest error rate. In the process of operation time prediction, the algorithm can select a basic model with better performance, effectively reduce the influence of a model with poorer performance and finally improve the prediction performance. Therefore, the regression integrated prediction method provided by the embodiment is superior to the existing single machine learning regression model.
In one embodiment, generating a candidate model combination comprises: acquiring a configured model set; and randomly selecting a preset number of models from the model set to form candidate model combinations.
Specifically, a preconfigured model set ModelList is obtained, which contains all models for combining. The server selects a preset number from the model set ModelList by adopting a random strategy, for example, ten models are selected to form a candidate model combination ModelSet, and the ModelSet belongs to the ModelList. That is, if a total of 5 candidate model combinations are generated to perform the integration training to obtain 5 candidate integrated models, the generated 5 candidate model combinations each include ten randomly selected models.
It should be understood that, although the steps in the flowchart of fig. 2 are shown in order as indicated by the arrows, the steps are not necessarily performed in order as indicated by the arrows. The steps are not performed in the exact order shown and described, and may be performed in other orders, unless explicitly stated otherwise. Moreover, at least a portion of the steps in fig. 2 may include multiple steps or multiple stages, which are not necessarily performed at the same time, but may be performed at different times, which are not necessarily performed in sequence, but may be performed in turn or alternately with other steps or at least a portion of the other steps or stages.
In one embodiment, as shown in fig. 6, there is provided a simulation runtime prediction apparatus, including: a data collection module 602, a model generation module 604, an integration training module 606, and an evaluation application module 608, wherein:
a data collection module 602, configured to collect operation parameters of the simulation application operation to obtain a model-entry operation parameter;
the model generation module 604 is configured to generate a candidate model combination and obtain a preset meta-model to be trained;
an integrated training module 606, configured to perform integrated training on each candidate model combination and the meta-model to be trained by using the in-mold operation parameters to obtain a candidate integrated model;
an evaluation application module 608 is configured to evaluate each candidate integration model, determine a simulation runtime prediction model, and predict a runtime of the simulation application using the runtime prediction model.
In one embodiment, the data collection module 602 is further configured to monitor and collect pre-operation parameters and runtime parameters of the simulation application by using the cloud computing node, so as to obtain the operation parameters of the simulation application; and selecting the characteristics of the operation parameters, and screening to obtain the mold-entering operation parameters.
In one embodiment, the data collection module 602 is further configured to perform feature selection on each operating parameter by using a variance selection method, so as to obtain feature importance of each operating parameter; and sorting the operation parameters according to the feature importance screening, and screening to obtain the in-mold operation parameters.
In one embodiment, the evaluation application module 608 is further configured to predict the running time of each candidate integration model by using the verification running parameter, respectively, to obtain an evaluation prediction time; determining the real running time corresponding to the verification running parameters, and calculating the average value of each real running time to obtain the average running time; obtaining the evaluation value of each candidate integration model according to the evaluation prediction time, the real running time and the average running time; and determining a simulation running time prediction model according to the evaluation value of each candidate integration model.
In one embodiment, the integrated training module 606 is further configured to train each model in the candidate model combination independently, with the input of the mold-entering operation parameter and the operation time corresponding to the mold-entering operation parameter as a target; respectively predicting the running time by using each model in the trained candidate model combination; fusing the operation time predicted by each model in the candidate model combination and the input operation parameters corresponding to the predicted operation time into an original training set to obtain a fused training set; and training the meta-model to be trained according to the fusion training set to obtain a trained candidate integrated model.
In one embodiment, the integrated training module 606 is further configured to train each model in the model set independently using K-fold cross validation.
In one embodiment, the model generation module 604 is further configured to obtain a configured set of models; and randomly selecting a preset number of models from the model set to form candidate model combinations.
For specific limitations of the simulation runtime prediction apparatus, reference may be made to the above limitations of the simulation runtime prediction method, which are not described herein again. The various modules in the simulation runtime prediction apparatus described above may be implemented in whole or in part by software, hardware, and combinations thereof. The modules can be embedded in a hardware form or independent from a processor in the computer device, and can also be stored in a memory in the computer device in a software form, so that the processor can call and execute operations corresponding to the modules.
In one embodiment, a computer device is provided, which may be a server, the internal structure of which may be as shown in fig. 7. The computer device includes a processor, a memory, and a network interface connected by a system bus. Wherein the processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device comprises a nonvolatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of an operating system and computer programs in the non-volatile storage medium. The database of the computer device is used for storing data such as operating parameters. The network interface of the computer device is used for communicating with an external terminal through a network connection. The computer program is executed by a processor to implement a method for ensemble learning based runtime prediction of a simulation of a complex system.
Those skilled in the art will appreciate that the architecture shown in fig. 7 is merely a block diagram of some of the structures associated with the disclosed aspects and is not intended to limit the computing devices to which the disclosed aspects apply, as particular computing devices may include more or less components than those shown, or may combine certain components, or have a different arrangement of components.
In one embodiment, a computer device is provided, comprising a memory and a processor, the memory having a computer program stored therein, the processor implementing the following steps when executing the computer program:
collecting operating parameters of simulation application operation to obtain mold-entering operating parameters;
generating a candidate model combination and acquiring a preset meta-model to be trained;
performing integrated training on each candidate model combination and the meta-model to be trained by using the in-mold operation parameters to obtain a candidate integrated model;
and evaluating each candidate integration model, determining a simulation run-time prediction model, and predicting the run time of the simulation application by using the run-time prediction model.
In one embodiment, the processor, when executing the computer program, further performs the steps of: monitoring and collecting the pre-operation parameters and the operation parameters of the simulation application by using the cloud computing nodes to obtain the operation parameters of the simulation application; and selecting the characteristics of the operation parameters, and screening to obtain the mold-entering operation parameters.
In one embodiment, the processor, when executing the computer program, further performs the steps of: respectively selecting the characteristics of each operation parameter by using a variance selection method to obtain the characteristic importance of each operation parameter; and sorting the operation parameters according to the feature importance screening, and screening to obtain the in-mold operation parameters.
In one embodiment, the processor, when executing the computer program, further performs the steps of: each candidate integration model respectively utilizes the verification operation parameters to predict the operation time to obtain the evaluation prediction time; determining the real running time corresponding to the verification running parameters, and calculating the average value of each real running time to obtain the average running time; obtaining the evaluation value of each candidate integration model according to the evaluation prediction time, the real running time and the average running time; and determining a simulation running time prediction model according to the evaluation value of each candidate integration model.
In one embodiment, the processor, when executing the computer program, further performs the steps of: respectively and independently training each model in the candidate model combination by taking the operation parameters of the mold entering as input and the operation time corresponding to the operation parameters of the mold entering as a target; respectively predicting the running time by using each model in the trained candidate model combination; fusing the operation time predicted by each model in the candidate model combination and the input operation parameters corresponding to the predicted operation time into an original training set to obtain a fused training set; and training the meta-model to be trained according to the fusion training set to obtain a trained candidate integrated model.
In one embodiment, the processor, when executing the computer program, further performs the steps of: each model in the independent training model combination is validated using K-fold cross-validation.
In one embodiment, the processor, when executing the computer program, further performs the steps of: acquiring a configured model set; and randomly selecting a preset number of models from the model set to form candidate model combinations.
In one embodiment, a computer-readable storage medium is provided, having a computer program stored thereon, which when executed by a processor, performs the steps of:
collecting operating parameters of simulation application operation to obtain mold-entering operating parameters;
generating a candidate model combination and acquiring a preset meta-model to be trained;
performing integrated training on each candidate model combination and the meta-model to be trained by using the in-mold operation parameters to obtain a candidate integrated model;
and evaluating each candidate integration model, determining a simulation run-time prediction model, and predicting the run time of the simulation application by using the run-time prediction model.
In one embodiment, the computer program when executed by the processor further performs the steps of: monitoring and collecting the pre-operation parameters and the operation parameters of the simulation application by using the cloud computing nodes to obtain the operation parameters of the simulation application; and selecting the characteristics of the operation parameters, and screening to obtain the mold-entering operation parameters.
In one embodiment, the computer program when executed by the processor further performs the steps of: respectively selecting the characteristics of each operation parameter by using a variance selection method to obtain the characteristic importance of each operation parameter; and sorting the operation parameters according to the feature importance screening, and screening to obtain the in-mold operation parameters.
In one embodiment, the computer program when executed by the processor further performs the steps of: each candidate integration model respectively utilizes the verification operation parameters to predict the operation time to obtain the evaluation prediction time; determining the real running time corresponding to the verification running parameters, and calculating the average value of each real running time to obtain the average running time; obtaining the evaluation value of each candidate integration model according to the evaluation prediction time, the real running time and the average running time; and determining a simulation running time prediction model according to the evaluation value of each candidate integration model.
In one embodiment, the computer program when executed by the processor further performs the steps of: respectively and independently training each model in the candidate model combination by taking the operation parameters of the mold entering as input and the operation time corresponding to the operation parameters of the mold entering as a target; respectively predicting the running time by using each model in the trained candidate model combination; fusing the operation time predicted by each model in the candidate model combination and the input operation parameters corresponding to the predicted operation time into an original training set to obtain a fused training set; and training the meta-model to be trained according to the fusion training set to obtain a trained candidate integrated model.
In one embodiment, the computer program when executed by the processor further performs the steps of: each model in the independent training model combination is validated using K-fold cross-validation.
In one embodiment, the computer program when executed by the processor further performs the steps of: acquiring a configured model set; and randomly selecting a preset number of models from the model set to form candidate model combinations.
It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above can be implemented by hardware instructions of a computer program, which can be stored in a non-volatile computer-readable storage medium, and when executed, can include the processes of the embodiments of the methods described above. Any reference to memory, storage, database or other medium used in the embodiments provided herein can include at least one of non-volatile and volatile memory. Non-volatile Memory may include Read-Only Memory (ROM), magnetic tape, floppy disk, flash Memory, optical storage, or the like. Volatile Memory can include Random Access Memory (RAM) or external cache Memory. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM), among others.
The technical features of the above embodiments can be arbitrarily combined, and for the sake of brevity, all possible combinations of the technical features in the above embodiments are not described, but should be considered as the scope of the present specification as long as there is no contradiction between the combinations of the technical features.
The above-mentioned embodiments only express several embodiments of the present application, and the description thereof is more specific and detailed, but not construed as limiting the scope of the invention. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the concept of the present application, which falls within the scope of protection of the present application. Therefore, the protection scope of the present patent shall be subject to the appended claims.
Claims (10)
1. An ensemble learning based runtime prediction method for simulation of a complex system, the method comprising:
collecting operating parameters of simulation application operation to obtain mold-entering operating parameters;
generating a candidate model combination and acquiring a preset meta-model to be trained;
performing integrated training on each candidate model combination and the meta-model to be trained by using the mold-entering operation parameters to obtain candidate integrated models;
and evaluating each candidate integration model, determining a simulation run-time prediction model, and predicting the run time of the simulation application by using the run-time prediction model.
2. The method of claim 1, wherein collecting operating parameters of the simulation application operation to obtain in-mold operating parameters comprises:
monitoring and collecting the pre-operation parameters and the operation parameters of the simulation application by using the cloud computing nodes to obtain the operation parameters of the simulation application;
and selecting the characteristics of the operation parameters, and screening to obtain the mold-entering operation parameters.
3. The method of claim 2, wherein the selecting the operating parameters and screening the operating parameters to obtain the in-mold operating parameters comprises:
respectively carrying out feature selection on each operation parameter by using a variance selection method to obtain the feature importance of each operation parameter;
and sorting the operation parameters according to the feature importance screening, and screening to obtain the in-mold operation parameters.
4. The method of claim 1, wherein said evaluating each of said candidate integration models to determine a simulation runtime prediction model comprises:
each candidate integration model respectively utilizes the verification operation parameters to predict the operation time to obtain the evaluation prediction time;
determining the real running time corresponding to the verification running parameters, and calculating the average value of the real running times to obtain the average running time;
obtaining an evaluation value of each candidate integration model according to the evaluation prediction time, the real running time and the average running time;
and determining a simulation running time prediction model according to the evaluation value of each candidate integration model.
5. The method according to claim 1, wherein the performing integrated training on each candidate model combination and the meta-model to be trained by using the in-mode operation parameters to obtain a candidate integrated model comprises:
respectively and independently training each model in the candidate model combination by taking the mold-entering operation parameter as input and the operation time corresponding to the mold-entering operation parameter as a target;
respectively predicting the running time by using each model in the trained candidate model combination;
fusing the operation time predicted by each model in the candidate model combination and the input operation parameters corresponding to the predicted operation time into an original training set to obtain a fused training set;
and training the meta-model to be trained according to the fusion training set to obtain a trained candidate integrated model.
6. The method of claim 5, wherein each model in the set of models is independently trained using K-fold cross validation.
7. The method of claim 1, wherein generating the candidate model combination comprises:
acquiring a configured model set;
and randomly selecting a preset number of models from the model set to form candidate model combinations.
8. An ensemble learning based runtime prediction apparatus for simulation of a complex system, the apparatus comprising:
the data collection module is used for collecting the running parameters of the running of the simulation application to obtain the in-mold running parameters;
the model generation module is used for generating a candidate model combination and acquiring a preset meta-model to be trained;
the integrated training module is used for performing integrated training on each candidate model combination and the meta-model to be trained by using the in-mode operation parameters to obtain a candidate integrated model;
and the evaluation application module is used for evaluating each candidate integration model, determining a simulation running time prediction model and predicting the running time of the simulation application by using the running time prediction model.
9. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor, when executing the computer program, implements the steps of the method of any of claims 1 to 7.
10. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the steps of the method of any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110330626.9A CN113158435B (en) | 2021-03-26 | 2021-03-26 | Complex system simulation running time prediction method and device based on ensemble learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110330626.9A CN113158435B (en) | 2021-03-26 | 2021-03-26 | Complex system simulation running time prediction method and device based on ensemble learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113158435A true CN113158435A (en) | 2021-07-23 |
| CN113158435B CN113158435B (en) | 2023-04-18 |
Family
ID=76885660
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110330626.9A Active CN113158435B (en) | 2021-03-26 | 2021-03-26 | Complex system simulation running time prediction method and device based on ensemble learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113158435B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114297946A (en) * | 2022-02-08 | 2022-04-08 | 无锡雪浪数制科技有限公司 | Industrial internet platform for realizing multidisciplinary simulation model order reduction |
| CN116560968A (en) * | 2023-07-11 | 2023-08-08 | 北京景行锐创软件有限公司 | Simulation calculation time prediction method, system and equipment based on machine learning |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180321980A1 (en) * | 2015-12-04 | 2018-11-08 | Cornell University | Execution time prediction for energy-efficient computer systems |
| US10402733B1 (en) * | 2015-06-17 | 2019-09-03 | EMC IP Holding Company LLC | Adaptive ensemble workload prediction model based on machine learning algorithms |
| US20190325307A1 (en) * | 2018-04-20 | 2019-10-24 | EMC IP Holding Company LLC | Estimation of resources utilized by deep learning applications |
| CN111258767A (en) * | 2020-01-22 | 2020-06-09 | 中国人民解放军国防科技大学 | Intelligent cloud computing resource allocation method and device for complex system simulation application |
| US20200349416A1 (en) * | 2019-05-05 | 2020-11-05 | Alibaba Group Holding Limited | Determining computer-executed ensemble model |
| US10841236B1 (en) * | 2018-03-30 | 2020-11-17 | Electronic Arts Inc. | Distributed computer task management of interrelated network computing tasks |
| US20210019189A1 (en) * | 2018-04-18 | 2021-01-21 | Landmark Graphics Corporation | Methods and systems to determine and optimize reservoir simulator performance in a cloud computing environment |
| CN112433806A (en) * | 2020-10-19 | 2021-03-02 | 湖南和信智仿信息科技有限公司 | Computing model structure and scheduling execution method for cloud simulation service |
-
2021
- 2021-03-26 CN CN202110330626.9A patent/CN113158435B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10402733B1 (en) * | 2015-06-17 | 2019-09-03 | EMC IP Holding Company LLC | Adaptive ensemble workload prediction model based on machine learning algorithms |
| US20180321980A1 (en) * | 2015-12-04 | 2018-11-08 | Cornell University | Execution time prediction for energy-efficient computer systems |
| US10841236B1 (en) * | 2018-03-30 | 2020-11-17 | Electronic Arts Inc. | Distributed computer task management of interrelated network computing tasks |
| US20210019189A1 (en) * | 2018-04-18 | 2021-01-21 | Landmark Graphics Corporation | Methods and systems to determine and optimize reservoir simulator performance in a cloud computing environment |
| US20190325307A1 (en) * | 2018-04-20 | 2019-10-24 | EMC IP Holding Company LLC | Estimation of resources utilized by deep learning applications |
| US20200349416A1 (en) * | 2019-05-05 | 2020-11-05 | Alibaba Group Holding Limited | Determining computer-executed ensemble model |
| CN111258767A (en) * | 2020-01-22 | 2020-06-09 | 中国人民解放军国防科技大学 | Intelligent cloud computing resource allocation method and device for complex system simulation application |
| CN112433806A (en) * | 2020-10-19 | 2021-03-02 | 湖南和信智仿信息科技有限公司 | Computing model structure and scheduling execution method for cloud simulation service |
Non-Patent Citations (4)
| Title |
|---|
| KAUR, G等: "An intelligent regressive ensemble approach for predicting resource usage in cloud computing", 《JOURNAL OF PARALLEL AND DISTRIBUTED COMPUTING》 * |
| MINA NAGHSHNEJAD等: "Adaptive Online Runtime Prediction to Improve HPC Applications Latency in Cloud", 《2018 IEEE 11TH INTERNATIONAL CONFERENCE ON CLOUD COMPUTING》 * |
| 孙天齐等: "云环境下Web应用带宽资源预测与管理", 《计算机应用》 * |
| 廖大强: "面向多目标的云计算资源调度算法", 《计算机***应用》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114297946A (en) * | 2022-02-08 | 2022-04-08 | 无锡雪浪数制科技有限公司 | Industrial internet platform for realizing multidisciplinary simulation model order reduction |
| CN114297946B (en) * | 2022-02-08 | 2023-03-24 | 无锡雪浪数制科技有限公司 | Industrial internet platform for realizing multidisciplinary simulation model order reduction |
| CN116560968A (en) * | 2023-07-11 | 2023-08-08 | 北京景行锐创软件有限公司 | Simulation calculation time prediction method, system and equipment based on machine learning |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113158435B (en) | 2023-04-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111258767B (en) | Cloud computing resource intelligent distribution method and device for complex system simulation application | |
| US10642642B2 (en) | Techniques to manage virtual classes for statistical tests | |
| Alipourfard et al. | {CherryPick}: Adaptively unearthing the best cloud configurations for big data analytics | |
| US9875142B2 (en) | System and method for efficient task scheduling in heterogeneous, distributed compute infrastructures via pervasive diagnosis | |
| Moreno et al. | An approach for characterizing workloads in *** cloud to derive realistic resource utilization models | |
| Ganapathi et al. | Statistics-driven workload modeling for the cloud | |
| US20070250630A1 (en) | Method and a system of generating and evaluating potential resource allocations for an application | |
| Soualhia et al. | Predicting scheduling failures in the cloud: A case study with *** clusters and hadoop on amazon EMR | |
| CN113158435B (en) | Complex system simulation running time prediction method and device based on ensemble learning | |
| Nadeem et al. | Modeling and predicting execution time of scientific workflows in the grid using radial basis function neural network | |
| JP2018005896A (en) | System, controller, method, and program for executing simulation job | |
| Gong et al. | Finding constant from change: Revisiting network performance aware optimizations on iaas clouds | |
| Avritzer et al. | A multivariate characterization and detection of software performance antipatterns | |
| US10248462B2 (en) | Management server which constructs a request load model for an object system, load estimation method thereof and storage medium for storing program | |
| JP3792879B2 (en) | Parallel execution system | |
| Poghosyan et al. | Managing cloud infrastructures by a multi-layer data analytics | |
| Ismaeel et al. | Real-time energy-conserving vm-provisioning framework for cloud-data centers | |
| Xue et al. | When computing meets heterogeneous cluster: Workload assignment in graph computation | |
| CN106874215B (en) | Serialized storage optimization method based on Spark operator | |
| Rehab et al. | Scalable massively parallel learning of multiple linear regression algorithm with MapReduce | |
| Ray et al. | A framework for analyzing resource allocation policies for multi-access edge computing | |
| KR20160044623A (en) | Load Balancing Method for a Linux Virtual Server | |
| CN111598390A (en) | Server high availability evaluation method, device, equipment and readable storage medium | |
| Singh et al. | Modular resource centric learning for workflow performance prediction | |
| Cammarota et al. | Pruning hardware evaluation space via correlation-driven application similarity analysis |
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 |