Research On Software Reliability Prediction Model Based On LSTM

Software reliability refers to the ability of software products not to cause system failure under specified conditions and within specified time.Software reliability prediction is the prediction of future software reliability based on the failure data observed and collected during testing,operation and maintenance.Software reliability prediction model is the basis of software reliability prediction,which can predict software failure behavior and guarantee software reliability by modeling software failure process.With the development of artificial intelligence technology,deep learning models have been applied in the field of software reliability prediction.Among them,the Long Short-Term Memory(LSTM)network has the ability of long-term memory,which is very suitable for solving the problem of time series prediction,and has achieved good results in the field of software reliability prediction.However,due to the lack of in-depth research on software failure data and the limitations of software reliability prediction model based on LSTM,the final prediction accuracy is difficult to reach the expectation.Based on the above problems,this thesis first proposes targeted solutions to the problems existing in software failure data,such as the problems of data set applicability difference and data shortage,and completes the construction of experimental data.Secondly,an optimization scheme is proposed for the software reliability prediction model based on LSTM from two aspects of network structure and model parameters,so as to improve the prediction effect of the model.The main research contents of this thesis are as follows:(1)To solve the problems of data set applicability difference and data shortage in the application of software failure data in the model,the corresponding solutions are put forward respectively.Software failure data sets include two types: cumulative failure data and failure interval data.The applicability of these two types of data sets is different in different models.To solve this problem,a comparative experiment of two data types is conducted on multiple models,and the experimental results show that cumulative failure data has a better fitting effect on the software reliability prediction model based on deep learning.In order to solve the problem of data shortage in software failure data sets,a method of data augmentation by data sampling is proposed,and the experimental data are constructed on three groups of public software failure data sets.(2)An improved flower pollination algorithm,IFPA,is proposed to optimize the parameters of software reliability prediction model based on LSTM.Firstly,the randomness of the initial population is enhanced by introducing chaotic mapping in the population initialization stage to improve the performance of the algorithm.Secondly,a new variant difference strategy is used in the local optimization stage to avoid falling into the local optimum.Then,in the global optimization stage,the convergence speed of the algorithm is improved by introducing the BAS algorithm.Finally,the IFPA algorithm proposed in this thesis is compared with the original algorithm to verify the optimization performance.(3)Combined with IFPA algorithm and attention mechanism,an optimized software reliability prediction model IA-LSTM is proposed to improve the prediction effect of the model.The IFPA algorithm is used to optimize the number of neurons in the hidden layer and the size of the sample training batch.The attention mechanism is used to give different weights to different input time steps of the model so as to realize effective processing of long time series input.To solve the problem that the length of model input sequence is difficult to be determined,a method is proposed to determine the length of model input sequence by using data autocorrelation analysis.Finally,the prediction performance of the software reliability prediction model proposed in this thesis is verified by experiments on three data sets,and the optimized model is compared with three existing models,MHPSO-LSTM,LSTM and RNN.The experimental results show that the software reliability prediction model proposed in this thesis has a better prediction effect than other models.