基于FrFT的组合预测模型在地铁基坑沉降监测中的应用

Application of combined prediction model based on FrFT in settlement monitoring of subway foundation pit

针对极限学习机(ELM)模型在地铁基坑沉降预测领域存在的参数选取困难问题,本文引入粒子群(PSO)算法对ELM模型的关键参数进行寻优,结合分数阶傅里叶变换(FrFT),提出一种基于FrFT与PSO-ELM模型的地铁基坑沉降组合预测模型。首先,采用FrFT对地铁基坑沉降数据进行多尺度分解,得到若干结构简单的子序列;其次,引入粒子群算法对ELM模型进行全局寻优,提升ELM预测性能,对于各子序列,使用提出的PSO-ELM模型分别进行建模预测;最后,叠加各子序列结果作为预测结果。使用某地铁基坑沉降监测数据进行预测实验,对比GM(1,1)模型、ELM网络模型、长短期记忆(LSTM)模型与本文组合预测模型的实验结果,证明本文提出的组合预测模型的预测精度较对比模型明显提升,从而验证了本文模型能够有效挖掘数据中存在的规律性与趋势性信息,同时解决参数选取困难的问题,为相关结构变形监测与预测提供了一定借鉴。

In view of the difficult parameter selection of the extreme learning machine(ELM)model in the field of subway foundation pit settlement prediction,this paper introduced the particle swarm optimization(PSO)algorithm to optimize the key parameters of the ELM model.Combined with fractional Fourier transform(FrFT),the paper proposed a combined prediction model of subway foundation pit settlement based on FrFT and PSO-ELM models.Firstly,FrFT was used to decompose the settlement data of the subway foundation pit on multiple scales,and several subsequences with simple structures were obtained.Secondly,the PSO algorithm was introduced to optimize the ELM model globally and improve the prediction performance of ELM.For each sub-sequence,the proposed PSO-ELM model was used for modeling and prediction,respectively.Finally,the results of superimposing each subsequence were considered predicted results.The settlement monitoring data of a subway foundation pit was used for the prediction experiment,and the experimental results of the GM(1,1)model,ELM network model,long-short term memory(LSTM)model,and the combined prediction model in this paper were compared.The results show that the combined prediction model proposed in this paper has significantly improved the prediction accuracy compared with other models,verifying that the model in this paper can effectively mine the regularity and trend information in the data,solve the problem of difficult parameter selection,and provide a reference for the monitoring and prediction of relevant structural deformation.