System Performance Evaluation Methods For Multi-source And Small Sample Test Data

With increasing complexity of the complex systems,the performance evaluation of complex systems has become an important part of the overall research and development process.Due to the long development periods and high testing expenses of complex systems,it is impossible to carry out large-scale actual performance test of complex systems,which makes the test data show the characteristics of small sample.And with the continuous development of computer simulation technology,the simulation data that can be used in the performance evaluation of complex systems presents the characteristics of multi-source.Therefore,this subject has carried out research on the system performance evaluation methods for multi-source and small sample test data,which provides methods and tools for the performance evaluation of the multi-source and small sample test data.Firstly,the prior information of multi-source is studied,and the prior information is classified according to different classification standards,and presented the representation methods of different types of prior information.The methods of preprocessing the prior information are studied.In view of the limitations of the traditional Bootstrap method,an improved Bootstrap method is proposed,which can expand the sample range to a certain extent,and avoid the deficiency of the traditional Bootstrap method that the sampling sample is limited to the original sample.Aiming at the fusion of the multi-source prior distribution,a prior distribution weighting method based on Kullback-Leibler divergence is proposed,which can effectively integrate prior information.Secondly,the relationship between the weight of posterior distribution and the weight of prior distribution is deduced according to Bayes theory.Analyzed the influence of different proposal distribution functions on Markov chain construction results and sampling efficiency when MCMC method is used to estimate the parameters of fusion posterior distribution,and proposed an adaptive construction method of proposal distribution,which can effectively reduce the influence of human factors on the proposal distribution,and improve the sampling efficiency of MCMC method for parameter estimation of fusion posterior distribution.Finally,a performance evaluation tool is designed and developed,the three-tier architecture model is used to design the architecture of this software,which has the characteristics of normalized structure,low coupling and strong scalability.This software includes the management of evaluation objects and evaluation indexes,indexes performance evaluation and comprehensive evaluation,and generation of the evaluation reports,and supports the features such as version control,log management,configuration file reading,which improves evaluation efficiency and work automation.The validity and practicability of this software is illustrated by an example of performance evaluation of a complex system.