Tên bài báo:

Comparison of GASVM and RSBPNN for Maximum Ground Settlement Prediction of Deep Foundation Pits

Tóm tắt:

The maximum ground settlement of deep foundation pit is related to mam factors, and there are complex nonlinear relationships. Traditional empirical formulas and numerical simulation methods are difficult to predict accurately, and yet the common system analysis methods such as support vector machine (SVM) and BP neural network (BPNN) have the problems of long prediction time, poor prediction precision and so on. In view of that, thừ paper proposed two approaches to improve the prediction performance of SVM and BPNN respectively. SVM was first optimized its parameters using genetic algorithm while built the genetic algorithm support vector machine (GASVM) prediction model. Titen, rough set was used to optimize BPNN by simplifying its input vector set, in this way the rough set BP neural network (RSBPNN) prediction model was built. Meanwhile, combined with 40 samples of metro station deep foundation pits in Shenzhen area, the above two prediction models were comtrastive studied. The study results indicated that the average mot mean square relative error (RMSRE) of GASVM prediction model is 2.59% compared with 5.94% of RSBPNN. Their values all meet the requirement of error control threshold (6%) for excellent prediction method, so they can be widely used in maximum ground settlement prediction of deep foundation pits. In addition, the RMSRE curve of GASVM has the lower fluctuation than RSBPNN, the residual values between predicted and measured values of GASVM range from 0.2 to 2.6mm while RSBPNN range from 0.3 to 4.5mm. Hence, it can be concluded that GASVM is a more precision and better prediction method than RSBPNN in deep foundation pit maximum ground settlement prediction.

Từ khóa: