基于GRA-LSTM与SARIMA组合模型的季节性时间序列预测  

Seasonal time series prediction based on combination model of GRA-LSTM and SARIMA

作　　者：罗广诚 郜家珏 蔡文学[1] LUO Guangcheng;GAO Jiajue;CAI Wenxue(Department of Electronic Commerce,South China University of Technology,Guangzhou 510000,China)

机构地区：[1]华南理工大学电子商务系,广州510000

出　　处：《智能计算机与应用》2021年第6期195-200,共6页Intelligent Computer and Applications

摘　　要：针对LSTM模型对季节性时间序列中的周期、趋势性变化不敏感的特点,提出将SARIMA模型与LSTM模型进行组合,以提高模型预测精度。该方法首先构建了以关键影响因素为非线性输入层和历史数据为线性输入层的多对一LSTM模型,将经过GRA法筛选的关键影响因素及历史数据输入到该模型中得到初步预测结果,使用SARIMA模型依据历史数据对季节性时间序列进行预测,提取预测结果中单位节点的比例序列,以实现对时间序列中周期、趋势信息的抽取,最后根据SARIMA模型中提取的单位节点比例对LSTM得到的初步预测结果进行修正,得到最终预测结果。实验选取某市民航春运客流量数据对组合模型精度进行验证,通过与支持向量机、GRA法、GRA-LSTM模型、SARIMA模型4种单模型进行比较,验证了组合模型对于季节性时间序列预测的优越性。In view of the insensitivity of LSTM model to periodic and trend changes in seasonal time series,this paper proposes to combine SARIMA model with LSTM model to improve the prediction accuracy of the model.The method firstly constructs a many-to-one LSTM model with key influencing factors as nonlinear input layer and historical data as linear input layer,and inputs the key influencing factors and historical data screened by GRA method into the model to obtain preliminary prediction results.then,the SARIMA model is used to predict seasonal time series according to historical data,and the proportion sequence of unit nodes in the prediction results is extracted to realize the extraction of cycle and trend information in time series.finally,the preliminary prediction results obtained by LSTM are corrected according to the proportion of unit nodes extracted in SARIMA model to obtain the final prediction The accuracy of the combined model is verified by selecting the passenger flow data of a city's civil aviation Spring Festival travel rush.Compared with four single models,such as support vector machine,GRA method,GRA-LSTM model and SARIMA model,the superiority of the combined model for seasonal time series prediction is verified.

关 键 词：LSTM模型 SARIMA模型 组合模型 季节性时间序列预测 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]