基于TCN与SVM组合模型的建筑沉降预测

Application of the TCN and SVM combined model in building settlement prediction

针对基坑周边建筑沉降序列随时间和因其他影响因素呈现的非平稳性变化特征,提出一种基于TCN与SVM的短期沉降组合预测方法。该方法利用串联组合的方式,将沉降序列看作自相关的时间序列和受外界影响的非线性部分的组合,再利用TCN与SVM模型对两部分分别进行预测,最后将两部分叠加后得到沉降数据预测结果。结果表明,TCN-SVM组合预测模型能够很好地跟踪建筑沉降变化,其预测精度比单一SVM模型提高10%~20%,可有效提高短期建筑沉降预测精度。

Aiming at the characteristics of non-stationary change of building settlement sequence around foundation pit with time and other factors,a combined short-term settlement prediction method based on TCN and SVM is proposed.This method uses the combination method,regards the subsidence sequence as the combination of the autocorrelation time series and the nonlinear part affected by the outside world,and then uses the TCN and SVM models to predict the two parts respectively,and finally superimposes the two parts to get prediction results.The results show that the TCN-SVM combined prediction model can track the changes of building settlement well,and its prediction accuracy is 10%~20%higher than that of the single SVM model,which effectively improves the accuracy of short-term building settlement prediction.