摘要
为提高边坡位移预测的精度,提出一种融合Gaussian-filter算法与LSTM预测算法的GF-LSTM混合预测模型,依托某铁路边坡工程监测数据进行验证分析。结果表明:GF-LSTM模型不仅能对原始监测数据集进行预处理,还能提供精准的预测结果;GF-LSTM预测模型可很好地反映边坡位移的上下波动,所得的预测值与实测值整体趋势贴合、相关性极高,R2分别为0.915、0.908,均大于0.900;由降噪前后对比可知:两测点R2分别增加了0.143、0.185,而MAE和MAPE分别降低了0.104与0.874%、0.246与0.755%,表明降噪处理后各测点的预测精度和预测误差得到改善。
In order to improve the accuracy of slope displacement prediction,it proposes a GF-LSTM hybrid prediction model integrating Gaussian-filter algorithm and LSTM prediction algorithm,and conducts an example verification analysis based on the monitoring data of railway slope engineering.The research shows that the GFLSTM model can not only pre-process the original monitoring data set,but also provide accurate prediction results;the GF-LSTM prediction model can well reflect the fluctuation of slope displacement,and the predicted values obtained fit the overall trend and correlation with the actual measured values,which R2 are 0.915 and 0.908 respectively,greater than 0.900;it can be seen from the comparison before and after noise reduction that of the two measuring points R2 increased by 0.143 and 0.185 respectively,while MAE and MAPE decreased by 0.104 and 0.874%,0.246 and 0.755%respectively.These data indicate that the prediction accuracy and prediction error of each measuring point were improved after noise reduction.
作者
景自强
JING Ziqiang(The Second Engineering Co.,Ltd.of China Railway 18th Bureau Group,Tangshan 063000,Hebei,China)
出处
《路基工程》
2023年第3期181-186,共6页
Subgrade Engineering
关键词
铁路边坡
混合预测模型
降噪处理
位移预测
长短期记忆网络
railway slope
hybrid prediction model
noise reduction processing
displacement prediction
Long and Short Term Memory Network(LSTM)