基于SATLSTM的Web系统老化趋势预测

为解决传统软件老化趋势预测模型泛化能力弱、预测准确度低的问题,根据老化数据的时序特性,提出一种基于自注意力机制的长短时记忆单元(Self-Attention-Long Short Term Memory,SATLST)循环神经网络老化趋势预测模型。将LSTM循环神经网络与自注意力机制融合,在深入挖掘老化数据潜在规律的同时,通过对不同输入数据分配不同权重的方式,加大对局部老化信息的关注度。应用加速寿命测试实验搭建软件老化测试平台,对Web服务器因内存泄漏引起的老化现象进行建模和预测。实验结果表明:相较于传统的ARIMA和BP神经网络模型,该模型训练结果与实际值接近,预测精度高,能准确预测Web系统软件老化趋势。

PREDICTION OF WEB SYSTEM AGING TREND BASED ON SATLSTM

In order to solve the problem of weak generalization ability and low prediction accuracy of traditional software aging trend prediction model,according to the timing characteristics of aging data,this paper proposes an aging trend prediction model based on self-attention mechanism SATLST recurrent neural network.This model combined the LSTM recurrent neural network with self-attention mechanism.While deepening the potential law of aging data,we paid more attention to local aging information by assigning different weights to different input data,and built a software aging test platform with accelerated life test experiment to model and predict the aging phenomenon caused by memory leak of Web server.The experimental results show that compared with the ARIMA and BP neural network models,the training results of our model are close to the actual value.It has high prediction accuracy,which can accurately predict the aging trend of Web system software.