Research On Aging Prediction Of Software System Based On Artificial Neural Network

Software system has produced a serious of phenomena about software performance decline in a long-running and uninterrupted operation process, such as the average responsive time becomes slow and even becomes unresponsive, the computational error rate rises, the data pocket losses, the rate of throughput drops down and so on. Due to a lot of memory is occupied or leaked, file lock is not released in time, data is not updated in time, rounding error is accumulated during computing and other reasons. It is known as software aging. Software aging will lead to economic losses, even lead to the people?s life safety. Thereforeit is timely and accurately to predict the trend of software development, and avoid unnecessary loss of software agingthrough a series of software rejuvenation strategy.The average responsive time of software system is an important indicator of software system performance.For the user, the average responsive time is the user’s personal experience. The thesis judges whether the software is in the state of aging by predicting the length of the average responsive time. Firstly based on BP neural network as the basis and the software aging prediction model is designed, and then through the AdaBoost algorithm and BP neural network were combined with each other, based on AdaBoost algorithm of BP neural network software aging prediction modelis designed, lastly based on the extreme learning machineof software aging prediction modelis designed.The experimental results show that based on AdaBoost algorithm of BP neural network prediction model and based on extreme learning machine prediction model can improve the inherent defects of BP neural network, during predicting the average responsive time they showed higher prediction accuracy, and can judge the software agingmore accurately.