基于长短期记忆网络的地热数据预测——以毛垭温泉泉水温度数据为例

Geothermal Data Prediction Based on Long Short-Term Memory Network:Take Water Temperature Data of Maoya Spring Water as an Example

地热异常与地震关系密切。地热数据是典型的时间序列数据,研究其变化规律是检测异常数据的前提。传统的时间序列分析方法主要以线性方法为主,其拟合精度有限。根据地热时间序列数据的特点,论文采用LSTM模型对毛垭温泉泉水温度4年观测数据进行建模,并将实验结果与传统的AR方法、ARMA方法进行了对比。实验结果表明,在毛垭温泉泉水温度数据集上,LSTM方法预测的均方根误差RSME明显小于AR和ARMA方法。论文研究为地震前兆数据预测拓宽了思路。

Geothermal anomalies are closely related to earthquakes.Geothermal data is a typical time series data,and the study of its variation law is the basis of detecting abnormal data.The traditional time series analysis method is mainly linear method,and its fitting accuracy is limited.According to the characteristics of geothermal time series data,we model the observation data of water temperature in Maoya hot spring in four years by the LSTM model,and compare its experimental results with that of the traditional AR method and ARMA method.The experimental results show that the Root Mean Square Error predicted by LSTM method is significantly smaller than that predicted by AR and ARMA methods.The research of this paper gives the new thinking of earthquake precursor data prediction.