Prediction Of Software Aging Trend Based On Particle Swarm Annealing Neural Network

Software aging is usually caused by the cumulative effect of software fault activation,which leads to system errors such as memory leaks and exhaustion of operating system resources,resulting in software failure or system downtime.Although anti-aging operation can reduce the system performance degradation caused by software aging and failure,when to implement the anti-aging operation remains to be studied.The frequency of software regeneration is critical,too frequent will prolong the downtime,too low frequency will not solve the aging problem in time,and good operation effect cannot be guaranteed.Therefore,the key point of when to adopt the regeneration strategy is whether to accurately predict the trend of software aging,according to the prediction results,the regeneration strategy is timely adopted,which ensures the reliability of the software to a certain extent.In order to accurately grasp the overall trend of software aging,combined with the non-linear,complex and changeable characteristics of aging data,the aging phenomenon of web server caused by memory leakage is modeled and predicted.First of all,by building a software aging experiment platform with memory leakage and applying accelerated life theory,and the memory variation,which can reflect the aging trend,is selected as the experimental data set.Secondly,because BP neural network is a complex nonlinear dynamic learning system,based on this characteristic,a BP neural network prediction model is established,and the network structure and algorithm flow of the prediction model are discussed.Thirdly,in view of the essential defect that the weights and thresholds of the BP neural network are set randomly,the New Particle Swarm Annealing Algorithm(NPSOSA)is introduced to further propose and establish the NPSOSA-BP prediction model.Finally,the simulation experiment was conducted in MATLAB 2016 a,different models use the same aging data during the experimental training process,through the comparative analysis among the models,the validity of the prediction model is proved.The experimental results show that the prediction model proposed in this paper utilizes the nonlinear and adaptive characteristics of BP neural network,the parallelism and global convergence of particle swarm optimization algorithm,probability jump and local convergence of simulated annealing algorithm,respectively.The evaluation and analysis of the prediction results of different models prove that the NPSOSA-BP neural network model has improved prediction accuracy and applicability compared to the traditional particle swarm optimization(PSO)and traditional particle swarm annealing algorithm(PSOSA)optimized BP neural network models,the validity of the method in this application field is verified.