基于LSTM的软件时间序列延迟预测仿真  

Software Time Series Delay Prediction Simulation Based on LSTM

作　　者：夏容[1] 江官星[1] XIA Rong;JIANG Guan-xing(Science and Technology College,NCHU,Gongqingcheng Jiangxi 332020,China)

机构地区：[1]南昌航空大学科技学院,江西共青城332020

出　　处：《计算机仿真》2021年第12期435-439,共5页Computer Simulation

基　　金：江西省教育厅科学技术研究项目(GJJ171519);江西省高等学校教学改革研究课题(JXJG-20-38-1)。

摘　　要：针对传统方法对软件时间序列延迟预测的预测准确率低,预测时间长、漏报率高的问题,提出基于LSTM的软件时间序列延迟预测方法。采用激活函数对LSTM(长短期记忆网络)中门控机制与隐藏层的权值系数进行计算,提取软件中的数据特征;利用支持向量机中的函数拟合方法,通过软件中提取的数据特征,构建软件时间序列延迟预测模型。最后将软件的时间序列放入预测模型中,利用核函数对模型进行计算,实现时间序列从低维空间到高维空间的映射,将低维空间的非线性问题转化为高维的线性问题,通过相应的拟合函数计算出结果,以此完成对软件时间序列延迟的预测。实验结果表明,运用该方法对软件的时间序列延迟进行预测,预测的时间短、准确率高、漏报率低。Aiming at the problems of low prediction accuracy, long prediction time and high false alarm rate of traditional methods for software time series delay prediction, a software time series delay prediction method based on LSTM is proposed. Activation function was applied to calculate the weight coefficients of gating mechanism and hidden layer in long-term and short-term memory network(LSTM) for extracting the data features in the software.According to the function fitting method of support vector machine and the data features extracted from the software,the delay prediction model of software time series was established. The time series of software were input into the prediction model to calculate the model, realizing the mapping of time series from low dimensional space to high dimensional space. The nonlinear problem in low dimensional space was transformed into a linear problem in high dimensional space. Based on the corresponding fitting function, the results were obtained to complete the prediction. The results show that the software time series delay predicted by this method has short prediction time, high accuracy and low missing report rate.

关 键 词：时间序列延迟 预测方法 门控机制 带状区域 

分 类 号：TP31[自动化与计算机技术—计算机软件与理论]