Probability Prediction Of Remaining Useful Life Of Software Systems

Running software systems for a long time will make the system run slower,and the use of multiple threads will cause the system memory to be insufficient and lead to lag,and eventually the system will fail or even crash,a phenomenon called software aging.Since the emergence of the concept of software aging,researchers have deepened their knowledge and theoretical knowledge of software aging,how to slow down software aging and how to avoid softening aging has become the main research content.For software aging software system,the traditional maintenance measures due to the technical lag is not enough to meet the needs of modern maintenance.With the continuous progress of scientific research,people found that the timely implementation of anti-aging operations can reduce the adverse effects caused by software aging.Software aging is a method of optimizing the operating environment,eliminating accumulated errors,and periodically updating the program to its original state.Although software failure prevention avoids the problem of accumulated errors,it requires periodic testing,which results in the loss of software system maintenance.If we can derive the optimal time to resist failure by predicting the Remaining Useful Life(RUL)of software systems,we will reduce the error of rejuvenation decision.Currently,the data-driven approach has become the mainstream research method in the field of remaining life prediction research.Although the prediction performance of this method is very good,it ignores the consideration of uncertainty in the prediction process and cannot obtain more accurate prediction information.Therefore,in this paper,on the basis of data-driven prediction,deep learning and uncertainty quantification are fully considered,while taking into account the software system in complex environmental conditions,the remaining life prediction technique is studied,and the main research work is as follows:(1)To address the problem of uncertainty in the prediction process,which is mostly not considered in the traditional deep learning research process,a method for predicting the probability of remaining life based on Bayesian Long Short-Term Memory(Bayesian LSTM)neural network is proposed through the understanding of the The LSTM network structure is improved by modeling the relationship between each input and the corresponding output as a Gaussian distribution through Bayesian linear regression,which is used to calculate confidence intervals,which can be used for uncertainty expression.Finally,experiments are conducted on the dataset to verify that the method has good prediction results.(2)To address the problem that traditional time-series prediction methods only model one-dimensional time series and it is difficult to effectively use other features,a DeepAR-based remaining life prediction method is proposed.The method can take into account other characteristics related to the prediction,perform trend prediction in the absence of data,and then use the proposed Bayesian long-and short-term network model to obtain the probabilistic predictive distribution of RUL.The effectiveness of the method is verified by conducting experiments on the dataset.(3)The development of a deep learning-based remaining lifetime prediction system for software systems is implemented,and the system framework is designed using PyQt5 to encapsulate the Bayesian long-and short-term memory neural network model,which enables the visualization of the prediction system.