基于LSTM⁃TCN的地下水位数据修复及应用  

Restoration and application of groundwater level data based on LSTM⁃TCN

作　　者：袁志洪 陈雨[1] Yuan Zhihong;Chen Yu(College of Electronics and Information Engineering,Sichuan University,Chengdu 610065,China)

机构地区：[1]四川大学电子信息学院,成都610065

出　　处：《现代计算机》2023年第8期20-26,共7页Modern Computer

摘　　要：准确预测地下水位是合理开发利用地下水资源的重要依据。由于数据采集过程中可能受到环境干扰或者采集装备发生故障等,导致数据缺失,进而影响模型的预测准确度。为了在含较多缺失值的地下水位数据上获得更准确的预测结果,首先,提出了基于长短时记忆网络(long short term memory,LSTM)-因果卷积网络(temporal convolu⁃tional network,TCN)的地下水位修复模型,通过学习原数据集中的时序特性和分布特点以改善数据集质量。然后,将多头注意力机制(multi⁃head attention mechanism,MA)与LSTM相结合进行地下水位预测,进一步提升LSTM模型的预测性能。最后,预测结果表明,使用LSTM⁃TCN方法修复后的数据集训练MA⁃LSTM模型,显著地提高了地下水位预测精度。Accurate prediction of groundwater level is an important basis for rational development and utilization of groundwater resources.Due to the possible interference of the environment or the failure of the acquisition equipment in the process of data acquisition,the data is missing,which affects the prediction accuracy of the model.In order to obtain more accurate prediction results on groundwater level data with more missing values,a groundwater level restoration model based on long short term memory(LSTM)temporal convolutional network(TCN)is proposed to improve the quality of data sets by learning the time series and distribution characteristics of the original data sets.Then,the multihead attention mechanism(MA)and LSTM are combined to predict the groundwater level to further improve the prediction performance of LSTM model.Finally,the prediction results show that using the LSTMTCN method to train the MALSTM model can significantly improve the prediction accuracy of groundwater level.

关 键 词：数据修复 生成对抗插补网络 因果卷积网络 地下水位预测 多头注意力机制 

分 类 号：P641.7[天文地球—地质矿产勘探]