摘要
传感器时序数据预测作为工业自动化和智能化的关键过程,对于自动化生产监督、风险预防和技术改进等具有重要意义。考虑到传统基于统计学的时序分析方法通用性弱、普通循环神经网络模型存在长期依赖的不足,针对工业设备温度、压力和电流强度等时序数据预测问题,提出了一种基于多变量分析的长短时记忆神经网络时序预测方法,该方法利用数据的远距离信息和多变量相关性,有效地提高了工业传感器时序数据预测的准确性。实验选取瑞典某公司的机械装载传感器数据用于训练和测试,通过与单变量长短时记忆模型以及其它主流时序预测算法比较,证明了该方法具备较好的预测性能和通用性。
The forecasting of sensor timing series is a key process for industrial automation and intelligentization,and is of great significance for automated production supervision,risk prevention and technological improvement. Considering the weak generality of traditional statistical-based time series analysis methods and the long-term dependence problem of common Recurrent Neural Network models,a method of Long Short-Term Memory Neural Network based on multivariable analysis is proposed to solve these problems,especially for forecasting temperature,pressure,and current intensity of industrial equipment. This method uses the long-term information of the data and multi-variable correlation,effectively improves the accuracy. The experiment selects a Swedish company＇s mechanical sensor dataset for training and testing,and compares the proposed method with the univariate long short-term memory model and other major temporal prediction algorithms. The experimental results showthat the proposed method has better prediction performance and versatility.
作者
易利容
王绍宇
殷丽丽
杨青
顾欣
YI Lirong;WANG Shaoyu;YIN Lili;YANG Qing;GU Xin(School of Computer Science and Technology,Donghua University,Shanghai 201620,China)
出处
《智能计算机与应用》
2018年第5期13-16,共4页
Intelligent Computer and Applications
基金
国家自然科学基金青年基金项目(61703092)
关键词
时间序列预测
长短时记忆神经网络
多变量分析
time series forecasting
Long Short-Term Memory Neural Network
multivariable analysis
