Cloud simulation computing resource prediction method, device and equipment based on ranking learning

Abstract

The application relates to a cloud simulation computing resource prediction method, device and equipment based on ranking learning, which comprises the following steps: acquiring a feature data set of a simulation application on a cloud computing node; the data in the feature data set comprises pre-operation parameters and runtime parameters of the simulation application; relevant feature data which significantly affects the running time of the simulation application is extracted from the feature data set based on a SHAP feature extraction method; the relevant feature data is input into a constructed ranking learning model to predict the priority ranking of different cloud computing resource usage; and the cloud computing resource usage with the highest ranking output by the ranking learning model is taken as an optimal cloud simulation computing resource prediction result. The relevant feature data which significantly affects the running time of the simulation application is extracted as model input by using the SHAP explainability-based feature extraction method, and the ranking learning model is used to predict the cloud computing resource usage with the highest ranking, so that the resource prediction performance is greatly improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of cloud simulation and relates to a cloud simulation computing resource prediction method, device and equipment based on ranking learning. BACKGROUND

[0002] A complex system is a whole composed of a large number of nodes that are functionally interrelated, interacted and complemented. The relationship between these systems is complex, and the interaction, mutual restriction and mutual influence are obvious. Simulation technology is one of the most important methods for researching complex systems at present. Due to the complexity, uncertainty and emergence of the overall behavior of the internal relationship of the complex system, the simulation application of the complex system has a large amount of calculation and a long running time, thereby putting forward high requirements on the computing resources and communication ability. The traditional serial simulation or distributed simulation on an ordinary network cluster has problems such as relatively solidified allocation of computing and communication resources and low simulation running efficiency, and it is difficult to meet the requirements of strong timeliness of complex system simulation large sample running.

[0003] Parallel discrete event simulation (PDES) technology supports high-performance parallel simulation running by cooperating multiple computing units and is commonly used to improve the execution efficiency of complex system simulation applications. Before the running of a complex system simulation application based on PDES, the number of computing resources (usually referring to the number of CPU cores) required for the application is usually specified, and the simulation task is divided into different CPU cores for parallel calculation. Cloud computing technology can realize the collaborative management and on-demand allocation of computing / storage resources, and therefore can provide an efficient resource allocation means for the running of complex system simulation applications.

[0004] In recent years, a large number of researchers have studied simulation resource prediction methods for cloud environment based on classification / regression machine learning technology. These methods collect simulation application feature data, use classification / regression algorithms to predict the running time of simulation applications under different computing resources, then rank the simulation computing resources based on the prediction results (simulation running time), and finally obtain the computing resource that can make the simulation application run the shortest time. However, in the process of implementing the present application, the inventors found that the foregoing traditional simulation resource prediction method has the technical problem of low resource prediction performance. SUMMARY

[0005] In view of the problems in the foregoing traditional method, the present application proposes a cloud simulation computing resource prediction method based on ranking learning, which can greatly improve the resource prediction performance, a cloud simulation computing resource prediction device based on ranking learning, a computer device and a computer readable storage medium.

[0006] To achieve the above purpose, the embodiments of the present application adopt the following technical solutions:

[0007] In one aspect, a cloud simulation computing resource prediction method based on ranking learning is provided, comprising the steps of:

[0008] obtaining a feature data set of a simulation application on a cloud computing node; the data in the feature data set includes pre-run parameters and run-time parameters of the simulation application;

[0009] extracting relevant feature data that significantly affects the running time of the simulation application from the feature data set based on a SHAP feature extraction method;

[0010] inputting the relevant feature data into a constructed ranking learning model to predict the priority ranking of different cloud computing resource usage;

[0011] ranking the cloud computing resource usage output by the ranking learning model as the optimal cloud simulation computing resource prediction result.

[0012] In another aspect, a cloud simulation computing resource prediction device based on ranking learning is also provided, comprising:

[0013] a data acquisition module for obtaining a feature data set of a simulation application on a cloud computing node; the data in the feature data set includes pre-run parameters and run-time parameters of the simulation application;

[0014] a feature extraction module for extracting relevant feature data that significantly affects the running time of the simulation application from the feature data set based on a SHAP feature extraction method;

[0015] a ranking prediction module for inputting the relevant feature data into a constructed ranking learning model to predict the priority ranking of different cloud computing resource usage;

[0016] a result output module for ranking the cloud computing resource usage output by the ranking learning model as the optimal cloud simulation computing resource prediction result.

[0017] In yet another aspect, a computer device is also provided, comprising a memory and a processor, the memory storing a computer program, and the processor implementing the steps of the cloud simulation computing resource prediction method based on ranking learning described above when executing the computer program.

[0018] In still another aspect, a computer readable storage medium is also provided, which stores a computer program, and the computer program implements the steps of the cloud simulation computing resource prediction method based on ranking learning described above when executed by a processor.

[0019] One of the above technical solutions has the following advantages and beneficial effects:

[0020] The cloud simulation computing resource prediction method, device and equipment based on ranking learning, after obtaining the pre-operation parameters and runtime parameters of the simulation application on the cloud computing node, extracts relevant feature data that significantly affects the simulation application runtime based on the SHAP feature extraction method, then inputs the relevant feature data into the constructed ranking learning model to predict the priority ranking of different cloud computing resource usage, and finally takes the cloud computing resource usage with the highest ranking as the optimal cloud simulation computing resource prediction result.

[0021] In this way, the SHAP explainability-based feature extraction method is used to quantitatively evaluate the importance of the factors affecting the runtime of the complex system simulation, that is, firstly, the factor set affecting the simulation application running efficiency is analyzed, and the static features before the simulation application running and the dynamic features during the simulation application running are extracted as the training sample feature data. Then, the important contribution degree of each feature is calculated based on the SHAP value, and the relevant feature data that significantly affects the simulation application runtime is extracted as the model input. The ranking learning model is constructed to predict the cloud computing resource usage with the highest ranking, thereby greatly improving the cloud computing resource prediction performance. BRIEF DESCRIPTION OF DRAWINGS

[0022] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or the prior art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0023] Figure 1 A flowchart of the cloud simulation computing resource prediction method based on ranking learning in one embodiment;

[0024] Figure 2 A flowchart of the cloud simulation computing resource prediction method based on ranking learning in another embodiment;

[0025] Figure 3 A flowchart of the extraction of relevant feature data in one embodiment;

[0026] Figure 4 A flowchart of the ranking prediction of cloud computing resource usage in one embodiment;

[0027] Figure 5 A flowchart of the cloud simulation computing resource prediction method based on ranking learning in one embodiment;

[0028] Figure 6 A module structure diagram of the cloud simulation computing resource prediction device based on ranking learning in one embodiment. DETAILED DESCRIPTION

[0029] In order to make the purposes, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and not intended to limit the present application.

[0030] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used in the description herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the present application.

[0031] It should be noted that the reference herein to "embodiments" means that a particular feature, structure, or characteristic described in connection with the embodiment can be included in at least one embodiment of the application. The appearance of the phrase in various places in the specification does not necessarily all refer to the same embodiment, nor does it necessarily refer to a separate or alternative embodiment, which is mutually exclusive from other embodiments.

[0032] Those skilled in the art can understand that the embodiments described herein can be combined with other embodiments. The term "and / or" used in the specification and claims of the present application means any combination of one or more of the associated listed items and all possible combinations, and includes these combinations.

[0033] In the research work, the inventors found that in the prediction process, the traditional resource prediction method is usually based on high-precision prediction indicators, such as root mean square error (RMSE) or mean absolute error (MAE) indicators, to construct a prediction model, and the prediction target is to minimize the difference between the predicted running time and the actual running time. However, the goal of the simulation resource prediction task is actually to predict the optimal number of computing resources, so as to ensure the shortest simulation application running time. Machine learning prediction models based on high-precision prediction indicators tend to result in a worse ranking. Therefore, in the simulation computing resource prediction task, compared with the method of predicting the simulation running time under different computing resources and performing re-ranking, the method of directly learning the ranking of resource usage can more deeply explain the relationship between different resource usage and simulation application running time.

[0034] Learning to Rank method takes the relative ranking between two samples in the dataset as feedback information, and ensures the minimization of the probability of incorrect ranking between sample pairs through iterative learning, so as to generate the best ranking of all samples. In the training process, the ranking learning model usually considers that each kind of incorrect ranking has the same error cost. However, in the execution process of the simulation application, more attention is often paid to the predicted number of computing resources ranked first and the prediction result is expected to make the simulation application run in the shortest time. Therefore, in order to solve the technical problem of low resource prediction performance in the traditional resource prediction method, the application provides a cloud simulation computing resource prediction method based on ranking learning. By collecting static feature data and dynamic monitoring data of the simulation application, the influence degree of each factor on the running efficiency of the simulation application is analyzed based on the SHAP interpretable feature extraction method, so as to extract the significant factors affecting the running efficiency of the simulation application. During this period, the cost loss function is improved and designed to increase more punishment when the ranking of the best resource is incorrect, so as to quickly and accurately predict the computing resource that can make the simulation application run in the shortest time.

[0035] The embodiments of the application will be described in detail below with reference to the accompanying drawings.

[0036] Cloud computing technology provides an efficient resource management method for complex system simulation applications. Compared with the traditional method of directly predicting the simulation application computing resource usage amount by using classification / regression technology, the method of learning the ranking of resource usage amount can more deeply explain the relationship between different resource usage amounts and the running efficiency of the simulation application. Therefore, the cloud simulation computing resource prediction method based on ranking learning provided in the application is designed to use the ranking learning method to predict the priority of the computing resource used before the simulation application runs. The goal of this method is to minimize the number of incorrect resource pairs, obtain the cloud computing resource usage amount ranked first, and thus improve the execution performance of the simulation application.

[0037] Please refer to Figure 1 In one embodiment, the application provides a cloud simulation computing resource prediction method based on ranking learning, including the following processing steps S12 to S18:

[0038] S12, obtain the feature data set of the simulation application on the cloud computing node; the data in the feature data set includes the pre-running parameters and the running time parameters of the simulation application.

[0039] It can be understood that in actual application, the resource monitor can be deployed on the cloud computing node to accurately monitor and collect real-time running information of the simulation application, such as pre-running parameters and running-time parameters of the simulation application. The pre-running parameters are determined before the simulation application is executed, and can include, for example, known parameters such as the number of cloud computing nodes for executing the simulation application, the number of simulation application entities, the look-ahead value, and the simulation time. The running-time parameters reflect the performance difference of the simulation application under different cloud computing resources, and such parameters can be collected during the execution of the simulation application, and can include, for example, parameters such as CPU usage, memory usage, network throughput, network delay, and file system usage. The data set composed of the collected pre-running parameters and running-time parameters will be used as the feature data set for the subsequent processing steps.

[0040] In some embodiments, the resource monitor can collect parameter information every set time (for example, 5 seconds) and store it in the deployed cloud application feature database. The specific parameters can be shown in Table 1.

[0041] Table 1

[0042]

[0043] S14, extracting relevant feature data that significantly affects the running time of the simulation application from the feature data set based on the SHAP feature extraction method.

[0044] It can be understood that SHAP is a unified method for explaining the output of a machine learning model. Based on the SHAP feature extraction method, the influence of the feature dimension of each data in the feature data set on the running efficiency of the simulation application can be analyzed, so as to determine the influence degree of various influence factors on the running efficiency of the simulation application. The specific processing process based on the SHAP feature extraction method can be understood in the same way as the existing SHAP calculation method in the art. Finally, the feature data that significantly affects the running time of the simulation application can be extracted from the feature data set according to the size of the influence of the feature dimension of each data on the running efficiency of the simulation application. These feature data are called relevant feature data. The number of extracted relevant feature data can be determined according to the determination threshold of the significant influence on the running time of the simulation application, and the determination threshold can be distributed and set for different simulation applications.

[0045] S16, inputting the relevant feature data into the constructed ranking learning model to predict the priority ranking of different cloud computing resource usage.

[0046] It can be understood that the purpose of the ranking learning model is to predict the priority of resources required for the next run using the last run data. The model uses the relative ranking between each pair of samples in the training sample set as feedback information, and minimizes the probability of misranking between sample pairs through iterative learning, thereby generating the optimal ranking of all samples. Therefore, the embodiment adopts the constructed ranking learning model to predict the overall ranking of simulation computing resources and the optimal number of resources required for the shortest running time of the simulation application. After inputting the extracted relevant feature data into the ranking learning model for iterative learning, the priority ranking of different cloud computing resource usage can be output, that is, the ranking of each cloud computing resource usage corresponding to the long-to-short (or short-to-long) ranking of the simulation application running time under each cloud computing resource usage.

[0047] S18, the cloud computing resource usage with the highest ranking output by the ranking learning model is taken as the optimal cloud simulation computing resource prediction result.

[0048] It can be understood that after the ranking of each cloud computing resource usage is predicted, the cloud computing resource usage with the highest ranking can be output as the optimal cloud simulation computing resource prediction result, under which the simulation application running time is the shortest, that is, the simulation application running efficiency is the highest.

[0049] The cloud simulation computing resource prediction method based on ranking learning described above, by obtaining the pre-run parameters and run-time parameters of the simulation application on the cloud computing node, extracts relevant feature data that significantly affects the running time of the simulation application based on the SHAP feature extraction method, and then inputs the relevant feature data into the constructed ranking learning model to predict the priority ranking of different cloud computing resource usage. Finally, the cloud computing resource usage with the highest ranking is taken as the optimal cloud simulation computing resource prediction result.

[0050] In this way, the SHAP explainable-based feature extraction method is used to quantitatively evaluate the importance of the factors affecting the running time of the complex system simulation, that is, first, the factor set affecting the running efficiency of the simulation application is analyzed, and the static features before the simulation application runs and the dynamic features during the simulation application runs are extracted as training sample feature data. Then, the important contribution of each feature is calculated based on the SHAP value, and the relevant feature data that significantly affects the running time of the simulation application is extracted as the model input, and the ranking learning model is constructed to predict the cloud computing resource usage with the highest ranking, thereby greatly improving the cloud computing resource prediction performance.

[0051] In one embodiment, as shown in Figure 2 the above method further comprises the step S13:

[0052] S13, pre-processing the data in the feature data set; the pre-processing includes data cleaning, error screening and data standardization.

[0053] It can be understood that after obtaining the required feature data set, in order to reduce the possibility of data bias, the data can also be first subjected to common data preprocessing operations such as data cleaning, error screening and data standardization, etc. to eliminate biased data.

[0054] By reducing the possibility of data bias through the above step S13, the prediction accuracy and prediction efficiency can be further improved.

[0055] In one embodiment, as shown in FIG. 1, regarding the above step S14, the specific processing steps S141-S143 can include: Figure 3

[0056] S141, based on the SHAP feature extraction method, analyze the influence degree of each feature dimension in the feature data set on the simulation application running time, and determine the correlation of each feature dimension on the simulation application running time;

[0057] S142, remove the feature data in the feature data set that is not related to the simulation application running time;

[0058] S143, select the relevant feature data that significantly affects the simulation application running time; the relevant feature data includes strongly correlated feature data and weakly correlated and non-redundant feature data.

[0059] Specifically, after analyzing the influence degree of each feature dimension in the feature data set on the simulation application running time based on the SHAP feature extraction method, the correlation of each feature data on the simulation application running time can be determined, the irrelevant features are removed, and the strongly correlated feature data and the weakly correlated but non-redundant feature data are selected, thereby minimizing the occurrence of errors and helping to establish a more accurate prediction model, i.e. improving the prediction accuracy of the ranking learning model.

[0060] Through the above processing steps, the occurrence of errors in the prediction process can be minimized and the prediction accuracy can be improved.

[0061] In one embodiment, the construction process of the ranking learning model can specifically include the following steps:

[0062] According to the ranking learning goal of the cloud computing resource usage, a feedback function of the ranking learning model is determined;

[0063] Based on the feedback function, a sample weight is respectively given to each sample pair in the feature data set and the weight distribution of the sample pair is determined;

[0064] Based on the feedback function and the weight distribution, an utility function and a loss function of the ranking learning model are established, and the construction of the ranking learning model is completed.

[0065] ​It can be understood that, regarding the construction process of the ranking learning model, firstly, the problem is modeled:

[0066] Considering that the ranking target is Set X, it is necessary to rank a group of simulation applications with the same simulation static parameters (except the number of cores used) i The best computing resource requirement of the simulation application in the cloud can be represented as P i = f(S i (x i ,y i )), wherein x i = (x1, x2, … x d ,) represents a d-dimensional feature vector of the simulation application, y i represents the running time of the simulation application, and f is a ranking function. The goal of the ranking learning model is to learn a ranking function f from the feature data set to correctly rank the number of cloud computing resources to be used. Therefore, the feedback function is defined as:

[0067]

[0068] For any pair of samples S j and S k in a group of simulation applications, when S j (x j ,y j ) > S k (x k ,y k ), it means that the simulation running time after using P j resources is greater than the simulation running time after using P k resources. When S j (x j ,y j ) = S k (x k ,y k ), it means that the simulation running time after using P j resources is equal to the simulation running time after using P k resources. When S j (x j ,y j ) < S k (x k ,y k ), it means that the simulation running time after using P j resources is less than the simulation running time after using P k resources. Based on this, the above ranking learning task is converted into a binary classification task of learning the relative order between two sample pairs. Based on this, a weight D is assigned to each sample pair, representing the importance of correctly judging the key sample pair. The distribution of D is defined as:

[0069]

[0070] Where c represents a positive constant that satisfies:

[0071]

[0072] Therefore, the goal of a ranking learning model is to learn a final ranking function H using positive feedback sample pairs with weights D. During the learning process, the probability of a ranking error occurring between the best ranking and the predicted ranking is called the ranking loss. Its utility function Func and loss function Rloss are defined as follows:

[0073]

[0074]

[0075] Where r represents the number of iterations, α r h represents the weights of the weak classifier obtained in the current iteration. r Represent a function with values ​​of 0-1, when α r >0 indicates h r The ranking effect is positive, meaning the accuracy exceeds 50%. Therefore, as long as an appropriate α is chosen in each weak learning process... r and h r This minimizes the sorting loss Rloss of the final sorting result. In this embodiment, α can be set by referring to existing literature in this field. r The weight value is defined as:

[0076] α r = 0.5ln(1+r(h) max ) / (1-r(h) max (6)

[0077] In one embodiment, such as Figure 4 As shown, step S16 above can specifically include the following steps S161 to S165:

[0078] S161, Initialize the weight distribution of sample pairs in the relevant feature data and set a positive constant for the weight distribution;

[0079] S162, In the current iteration of the ranking learning model, the maximization of the utility function is taken as the learning objective. Weak learning iterative training is performed using the weight distribution of the current iteration to generate multiple weak classifiers and calculate the weights of the weak classifiers obtained in the current iteration.

[0080] S163, obtaining the ranking result of the current iteration according to the weight of the weak classifier obtained in the current iteration and the weak classifier;

[0081] S164, entering the next iteration after updating the weight distribution of the sample pair according to the loss function of the ranking learning model; the processing flow of the next iteration is the same as that of the current iteration;

[0082] S165, when the comprehensive ranking obtained in the continuous N iterations does not change, ending the iteration training of the ranking learning model and outputting the comprehensive ranking as the priority ranking of different cloud computing resource usage. N is a positive integer not less than 2.

[0083] Specifically, the algorithm steps of the ranking learning model can be shown in the following prediction algorithm flow:

[0084]

[0085] In the prediction algorithm, S(X, Y) represents the input simulation application features and label data, and D represents the initialized sample pair weight distribution.

[0086] 1) First, initialize D1=D, and set the value of c in formula (2) to make the initial D value in each sample pair the same.

[0087] 2) In the iteration process, maximize the utility function r(h) as the learning goal, and use D r to perform weak learning iteration training to generate multiple weak learners h r , and obtain the value of a r by calculation. Finally, the ranking result is calculated as

[0088] Where R represents the number of iterations, and H(x) represents the ranking score. The larger the score, the higher the ranking of the sample.

[0089] 3) According to the formula:

[0090]

[0091] update the D distribution value, where Rloss is the minimum objective function value in the current iteration, which is calculated using formula (5).

[0092] 4) Repeat steps 2) and 3) until the comprehensive ranking order obtained in the continuous N iterations does not change, which is taken as the final prediction ranking and the optimal order of resource usage is output. N can be 2 or 3, which can be selected according to actual needs.

[0093] In one embodiment, further, the loss function of the ranking learning model is:

[0094]

[0095] Among them, S j (x j ,y j ) and S k (x k ,y k ) represents a sample pair, ε(x) j,k ) represents the penalty factor, D(S) j (x j ,y j ),S k (x k ,y k )) represents the weight distribution of sample pairs, r represents the number of iterations, and α r h represents the weights of the weak classifier obtained in the current iteration. r This represents a function with values ​​between 0 and 1.

[0096] Understandable, such as Figure 5 The diagram shows the flowchart of the above method, where the loss function is the Hinge loss function and concat represents the loss function. To make the ranking learning model more suitable for resource prediction problems in simulation applications, it is necessary to consider the following: Assume the optimal resource ranking required for a single simulation application run is a kernel of Correct ranking = {4, 2, 3, 1}. There are two predicted ranking results: Ranking 1 = kernel {2, 4, 3, 1} and Ranking 2 = kernel {4, 2, 1, 3}. It can be seen that both Ranking 1 and Ranking 2 are incorrect. However, in simulation practice, the resource usage that minimizes the simulation application's runtime is more important, i.e., the resource quantity ranked first. Therefore, the cost of incorrect ranking in Ranking 1 is significantly higher than the cost of incorrect ranking in Ranking 2, because the ranking result in Ranking 2 can also guarantee the shortest simulation application runtime. In other words, a larger penalty should be added based on Ranking 1. However, the traditional Rankboost ranking learning method does not consider this and assumes that the costs of incorrect ranking in Ranking 1 and Ranking 2 are equal. Based on this, in this embodiment, a penalty factor ε(x) is added. j,k To ensure that more penalties are added when the ranking of the best resource is incorrect, the loss function in formula (5) above is transformed into the final loss function (8) above.

[0097] By using the loss function shown in formula (8), the ranking cost in the simulation calculation resource ranking process is optimized. Compared with traditional machine learning methods such as linear regression, multilayer perceptron, regression tree and random forest, the accuracy of ranking can be significantly improved, and the error rate can be reduced.

[0098] In one embodiment, the calculation process of the penalty factor can specifically include the following steps:

[0099] According to all cloud computing resource requirements in the simulation application group, a correct ranking is created, MRR is used as the evaluation standard of the ranking performance, and the score MRR of the correct ranking is calculated;

[0100] For each sample pair in the simulation application group, the position of the module pair is exchanged, and a new ranking score MRR is calculated;

[0101] The average drop ratio of the ranking score MRR of all sample pairs after exchange to the score MRR of the correct ranking is calculated, and the average drop ratio is used as the penalty factor.

[0102] It can be understood that the key problem of simulation calculation resource prediction based on ranking learning is the definition of the penalty factor. In this embodiment, a heuristic method is used to calculate the value of ε(x j,k ).

[0103]

[0104]

[0105] Firstly, the algorithm flow creates a correct ranking according to all calculation resource requirements in the simulation application group, and uses MRR as the evaluation standard of the ranking performance and calculates the optimal score MRR best (see formula 9).

[0106] Secondly, for each sample pair Q={S j (x j ,y j ),S k (x k ,y k )} in the simulation application group, the position of the module pair is exchanged, and a new ranking score MRR j,k is calculated.

[0107] Finally, the average drop ratio of the MRR value of all sample pairs after exchange to the MRR value of the correct ranking is calculated, and it is used as the value of the penalty factor ε(x j,k ).

[0108] In some embodiments, the present specification also gives the evaluation index for testing the above-mentioned prediction method:

[0109] To verify the accuracy of the ranking learning prediction method, two indicators are considered: Mean Reciprocal Rank (MRR) and Normalized Discounted Cumulative Gain (NDCG). The formal definitions of the indicator metrics are as follows:

[0110] (1) MRR focuses on the position of the first relevant element in the ranking result. The larger the position of the first relevant element, the larger the MRR. The specific calculation is as follows:

[0111]

[0112] Where K represents the number of samples, rank i represents the position of the best resource in the predicted computing resource ranking list for the ith sample.

[0113] (2) NDCG considers that the earlier the sample in the ranking, the more valuable it is, and the greater the impact on the evaluation result. The specific calculation is as follows:

[0114]

[0115] Where Z n is a generalization factor, so that the value range of NDCG is limited to [0, 1], c(i) represents the relevance level of the sample ranked in the ith position, and 2 c(i) -1 is the evaluation gain value of the sample. log(1+i) represents the discount weight of the document ranking position, i.e. the smaller i represents the earlier the ranking position. Finally, NDCG@n adds up the evaluation gain values of the first n documents in the ranking to obtain the performance evaluation value of the ranking model.

[0116] It should be understood that although Figures 1 to 4 the steps in the flowchart are displayed in sequence according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this article, the execution of these steps is not strictly limited in order, and these steps can be executed in other orders. Moreover Figures 1 to 4 at least part of the steps of may include multiple sub-steps or multiple stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution order of these sub-steps or stages is not necessarily sequential, but can be executed in rotation or alternation with other steps or sub-steps or stages of other steps.

[0117] Please refer to Figure 6In one embodiment, a cloud simulation computing resource prediction device 100 based on ranking learning is provided, comprising a data acquisition module 11, a feature extraction module 13, a ranking prediction module 15, and a result output module 17. The data acquisition module 11 is configured to acquire a feature data set of a simulation application on a cloud computing node; the data in the feature data set includes pre-run parameters and runtime parameters of the simulation application. The feature extraction module 13 is configured to extract relevant feature data that significantly affects the running time of the simulation application from the feature data set based on a SHAP feature extraction method. The ranking prediction module 15 is configured to input the relevant feature data into a constructed ranking learning model to predict the priority ranking of different cloud computing resource usage. The result output module 17 is configured to output the cloud computing resource usage with the highest ranking from the ranking learning model as the optimal cloud simulation computing resource prediction result.

[0118] The cloud simulation computing resource prediction device 100 based on ranking learning described above, through the cooperation of each module, after acquiring the pre-run parameters and runtime parameters of the simulation application on the cloud computing node, extracts relevant feature data that significantly affects the running time of the simulation application based on the SHAP feature extraction method, and then inputs the relevant feature data into the constructed ranking learning model to predict the priority ranking of different cloud computing resource usage, and finally outputs the cloud computing resource usage with the highest ranking as the optimal cloud simulation computing resource prediction result.

[0119] In this way, the SHAP explainable-based feature extraction method is used to quantitatively evaluate the importance of the factors affecting the running time of the complex system simulation, that is, first, the factor set affecting the running efficiency of the simulation application is analyzed, and the static features before the simulation application runs and the dynamic features during the simulation application runs are extracted as training sample feature data. Then, the SHAP value is used to calculate the important contribution of each feature and extract relevant feature data that significantly affects the running time of the simulation application as model input, and the ranking learning model is used to predict the cloud computing resource usage with the highest ranking, thereby greatly improving the cloud computing resource prediction performance.

[0120] In one embodiment, the cloud simulation computing resource prediction device 100 based on ranking learning described above can also be used to implement the steps or sub-steps added in other embodiments of the cloud simulation computing resource prediction method based on ranking learning.

[0121] As to the specific limitation of the cloud simulation computing resource prediction apparatus 100 based on ranking learning, reference can be made to the corresponding limitation of the cloud simulation computing resource prediction method based on ranking learning in the foregoing, which will not be repeated here. Each module in the cloud simulation computing resource prediction apparatus 100 based on ranking learning can be realized by software, hardware, or a combination thereof, in whole or in part. Each module can be embedded in or independent of a device with a specific data processing function in hardware, or stored in the memory of the foregoing device in software, so as to be called and executed by the processor to perform the operation corresponding to each module. The foregoing device can be, but is not limited to, various types of data processing devices in the prior art.

[0122] In one embodiment, a computer device is also provided, including a memory and a processor, the memory storing a computer program, and the processor implementing the following processing steps when executing the computer program: obtaining a feature data set of a simulation application on a cloud computing node; the data in the feature data set including pre-run parameters and runtime parameters of the simulation application; extracting relevant feature data that significantly affects the running time of the simulation application from the feature data set based on a SHAP feature extraction method; inputting the relevant feature data into a constructed ranking learning model to predict the priority ranking of different cloud computing resource usage; and taking the cloud computing resource usage with the highest ranking output by the ranking learning model as the optimal cloud simulation computing resource prediction result.

[0123] It can be understood that, in addition to the memory and the processor mentioned above, the computer device also includes other software and hardware components not listed in the present specification, which can be determined according to the specific data processing device model in different application scenarios, and will not be listed in detail in the present specification.

[0124] In one embodiment, the processor executing the computer program can also implement the steps or sub-steps added in the embodiments of the cloud simulation computing resource prediction method based on ranking learning.

[0125] In one embodiment, a computer readable storage medium is also provided, which stores a computer program, and the computer program is executed by a processor to implement the following processing steps: obtaining a feature data set of a simulation application on a cloud computing node; the data in the feature data set including pre-run parameters and runtime parameters of the simulation application; extracting relevant feature data that significantly affects the running time of the simulation application from the feature data set based on a SHAP feature extraction method; inputting the relevant feature data into a constructed ranking learning model to predict the priority ranking of different cloud computing resource usage; and taking the cloud computing resource usage with the highest ranking output by the ranking learning model as the optimal cloud simulation computing resource prediction result.

[0126] In one embodiment, the computer program, when executed by the processor, further implements the steps or sub-steps added in each of the above-mentioned cloud simulation computing resource prediction methods based on ranking learning.

[0127] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiments can be completed by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer readable storage medium. When the computer program is executed, the processes of each of the above-mentioned embodiments can be included. Any reference to memory, storage, database or other medium used in each of the embodiments provided in the present application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus dynamic random access memory (Rambus DRAM, RDRAM for short) and interface dynamic random access memory (DRDRAM).

[0128] Each of the technical features of the above embodiments can be combined arbitrarily. In order to make the description simple, not all possible combinations of each technical feature in the above embodiments are described, however, as long as the combinations of the technical features do not exist contradictory, they should be considered as the scope of the present application. The above embodiments only express several implementation manners of the present application, and the description is more specific and detailed, but it should not be understood as a limitation on the scope of the patent. It should be pointed out that for a person of ordinary skill in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which all belong to the protection scope of the present application. Therefore, the patent protection scope of the present application should be subject to the appended claims.

Claims

1. A cloud simulation computing resource prediction method based on ranking learning, characterized in that, The method comprises the steps of: obtaining a feature data set of a simulation application on a cloud computing node; the data in the feature data set comprises pre-run parameters and run-time parameters of the simulation application; extracting relevant feature data that significantly affects the running time of the simulation application from the feature data set based on a SHAP feature extraction method; inputting the relevant feature data into a constructed ranking learning model to predict the priority ranking of different cloud computing resource usage; ranking the cloud computing resource usage output by the ranking learning model as the optimal cloud simulation computing resource prediction result; the loss function of the ranking learning model is: wherein, and denotes a sample pair, denotes a penalty factor, denotes a weight distribution of the sample pair, r denotes the number of iterations, denotes the weight of the weak classifier obtained in the current iteration, denotes a function with 0-1 values, denotes a d-dimensional feature vector for a simulation application, denotes the running time of the simulation application; and the calculation of the penalty factor comprises: create a correct ranking according to all cloud computing resource requirements in the simulation application group, use MRR as the evaluation standard of ranking performance, and calculate the score MRR of the correct ranking; for each sample pair in the simulation application group, exchange the positions of the module pairs and calculate the new ranking score MRR; calculate the average drop ratio of the ranking score MRR of all sample pairs after exchange relative to the score MRR of the correct ranking, and take the average drop ratio as the penalty factor. 2.The ranking learning based cloud simulation computing resource prediction method according to claim 1, characterized in that, The step of extracting relevant feature data that significantly affects the running time of the simulation application from the feature data set based on a SHAP feature extraction method comprises: analyze the influence degree of each feature dimension in the feature data set on the running time of the simulation application based on the SHAP feature extraction method, determine the correlation of each feature dimension on the running time of the simulation application; remove the feature data in the feature data set that is not related to the running time of the simulation application; select the relevant feature data that significantly affects the running time of the simulation application; the relevant feature data comprises strongly correlated feature data and weakly correlated and non-redundant feature data. 3.The ranking learning based cloud simulation computing resource prediction method according to claim 1 or 2, characterized in that, The construction process of the ranking learning model comprises: determine the feedback function of the ranking learning model according to the ranking learning goal of the cloud computing resource usage; based on the feedback function, respectively give a sample weight to each sample pair in the feature data set and determine the weight distribution of the sample pair; based on the feedback function and the weight distribution, establish the utility function and the loss function of the ranking learning model, and complete the construction of the ranking learning model. 4.The ranking learning based cloud simulation computing resource prediction method of claim 3, wherein, The step of inputting the relevant feature data into the constructed ranking learning model to predict the priority ranking of different cloud computing resource usage comprises: initialize the weight distribution of the sample pair in the relevant feature data and set the positive constant of the weight distribution; in the current iteration process of the ranking learning model, maximize the utility function as the learning goal, perform weak learning iteration training with the weight distribution of the current iteration, generate a plurality of weak classifiers and calculate the weight of the weak classifier obtained in the current iteration; calculate the ranking result of the current iteration according to the weight of the weak classifier obtained in the current iteration and the weak classifier; after updating the weight distribution of the sample pair according to the loss function of the ranking learning model, enter the next iteration; the processing flow of the next iteration is the same as that of the current iteration. When the comprehensive ranking obtained in the continuous N times of iterations does not change, the iterative training of the ranking learning model is ended, and the comprehensive ranking is output as the priority ranking of different cloud computing resource usage; N is a positive integer not less than 2.

5. The ranking-learning based cloud simulation computing resource prediction method according to claim 1, wherein, The method further comprises: The method further comprises: 6.A cloud simulation computing resource prediction device based on ranking learning, characterized in that, The method further comprises: The method further comprises: The data acquisition module is configured to acquire a feature data set of the simulation application on the cloud computing node. The data in the feature data set comprises pre-run parameters and run-time parameters of the simulation application. The feature extraction module is configured to extract relevant feature data that significantly affects the running time of the simulation application from the feature data set based on a SHAP feature extraction method. The ranking prediction module is configured to input the relevant feature data into a constructed ranking learning model to predict a priority ranking of different cloud computing resource usage. The result output module is configured to output the cloud computing resource usage with the highest ranking output by the ranking learning model as a prediction result of the optimal cloud simulation computing resource. wherein, and denotes a sample pair, denotes a penalty factor, denotes a weight distribution of the sample pair, r denotes the number of iterations, denotes the weight of the weak classifier obtained in the current iteration, denotes a function with 0-1 values, denotes a d-dimensional feature vector for a simulation application, denotes the running time of the simulation application; and the penalty factor is calculated by: The loss function of the ranking learning model is: According to all cloud computing resource requirements in the simulation application group, a correct ranking is created, MRR is used as a ranking performance evaluation standard, and the score MRR of the correct ranking is calculated. For each sample pair in the simulation application group, the exchange module exchanges the positions of the pair and calculates a new ranking score MRR. 7.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is configured to perform the method according to any one of claims 1-6 when the computer program is executed by the processor. The average drop ratio of the ranking score MRR of all sample pairs after the exchange relative to the score MRR of the correct ranking is calculated, and the average drop ratio is used as the penalty factor.

8. A computer-readable storage medium having stored thereon a computer program, characterized in that, The processor executes the computer program to implement the steps of the cloud simulation computing resource prediction method based on ranking learning according to any one of claims 1 to 5. The computer program is executed by the processor to implement the steps of the cloud simulation computing resource prediction method based on ranking learning according to any one of claims 1 to 5.

Images