基于模糊理论与SVM的边坡地震失稳规模贝叶斯网络评估方法

Fuzzy theory-and SVM-based Bayesian network assessment method for slope seismic instability scale

由于历史边坡样本数量不足,使用传统机器学习方法进行边坡稳定性评估往往会出现过拟合问题。以贝叶斯网络为框架,结合模糊理论与支持向量机,建立一种边坡地震失稳规模评估新方法。采用模糊理论求解贝叶斯网络参数的先验分布,同时采用支持向量机求解贝叶斯网络参数的实际样本潜在分布;利用贝叶斯估计方法将先验分布与实际样本潜在分布结合,以得到既满足基本规律又体现样本非线性与随机性的边坡地震失稳规模贝叶斯网络。利用建立好的贝叶斯网络对32个实际边坡样本进行失稳规模评估,正确率为81.25%。通过与基于先验知识的贝叶斯网络、基于实际样本的支持向量机进行对比,可以看出,提出的方法解决了样本不足带来的过拟合问题,正确率有很大提升。并且,该方法在边坡属性不完整的情况下也能对失稳规模作出有效评估。

Due to the insufficiency of historical slope samples,there are frequently over-fitting problems when using traditional machine learning method to assess the stability of slope.Based on Bayesian network,this paper combines fuzzy theory with support vector machine to propose a new method for slope seismic instability scale assessment.The method utilizes fuzzy theory to solve the prior distribution of the Bayesian network parameter.Meanwhile,support vector machine is applied to solve the actual sample potential distribution of the Bayesian network parameter.Then Bayesian estimation method is used to combine the prior distribution with the actual sample potential distribution to obtain the Bayesian network of slope instability scale.This Bayesian network satisfies both the basic law and the sample nonlinearity and randomness.Using the established Bayesian network to evaluate the instability scale of 32 actual slope samples,the correct rate is 81.25%.By comparing with Bayesian network based on prior knowledge and support vector machine based on actual sample,it can be seen that the proposed method solves the over-fitting problem caused by insufficiency of samples,and the accuracy is greatly improved.Moreover,this method can also effectively assess the instability scale when the slope properties are incomplete.