利用Prophet模型进行地下水位异常识别初探  

Preliminary Study on Groundwater Level Anomaly Identification Using the Prophet Model

作　　者：李永生 周晨 张思萌 石伟 年华 LI Yongsheng;ZHOU Chen;ZHANG Simeng;SHI Wei;NIAN Hua(Heilongjiang Earthquake Agency,24 Hongxiang Road,Harbin 150090,China)

机构地区：[1]黑龙江省地震局,哈尔滨市150090

出　　处：《大地测量与地球动力学》2025年第3期313-318,共6页Journal of Geodesy and Geodynamics

基　　金：吉林长白山火山国家野外科学观测研究站课题(NORSCBS21-09);黑龙江省地震局火山活动性研究创新团队项目;中国地震局地震科技星火计划(XH20017)。

摘　　要：针对地下水位数据的复杂特性(包括非线性趋势、季节性波动和随机扰动),引入Facebook开发的Prophet时间序列预测模型,旨在利用其非线性趋势捕捉、季节性波动解析及对异常值和数据缺失的灵活应对能力,显著提升地下水位异常识别的准确性。黑龙江省绥化市北林区地震台观测数据表明,Prophet模型在捕捉时间序列动态特征上表现优越,能有效识别异常。模型调整后具有高拟合精度和高预测能力,预测误差低,决定系数高。此外,模型在地震预测中能识别出与地震相关的水位异常,可为地震前兆研究提供新视角。本文结果表明Prophet模型在处理复杂时间序列数据时具有可行性,可为地震预测提供新工具。We focus on the complex characteristics of groundwater level data,including nonlinear trends,seasonal fluctuations,and random disturbances,and introduce the Prophet time series prediction model developed by Facebook.The aim is to use its nonlinear trend capture,seasonal fluctuation analysis,and flexible response ability to outliers and data missing to significantly improve the accuracy of groundwater level anomaly identification.Through observation data from Beilin district seismic station in Suihua city,Heilongjiang province,it is shown that the Prophet model performs well in capturing dynamic characteristics of time series data and can effectively identify anomalies.The high fitting accuracy and predictive ability of adjusted model have been confirmed,with low prediction error and high determination coefficient.In addition,the model identifies water level anomalies related to earthquakes in earthquake prediction,providing a new perspective for earthquake precursor research.This study demonstrates the effectiveness of Prophet model in processing complex time series data,providing a new tool for earthquake prediction.

关 键 词：地下水位异常识别 时间序列预测 Prophet模型 地震预测 

分 类 号：P315[天文地球—地震学]