基于粒子群BP网络混合算法的边坡稳定性评价

Slope stability evaluation based on hybrid algorithm of particle swarm optimization and BP neural network

边坡稳定性评价与预测具有高度非线性和不确定性特征,难以用准确的数学模型表达。选取多个边坡工程实例构成学习样本集,以土体重度、内摩擦角、黏聚力、坡角、坡高、孔隙压力比6个主要影响因素作为土坡稳定性的评价判别指标;然后采用粒子群算法优化BP神经网络模型,实现混合算法,在保持BP网络算法误差反向传播修正权值特点的同时,将网络权值和阈值粒子化,利用粒子群算法的全局搜索性实现网络权值和阈值的更新,从而加快收敛速度和提高收敛精度,避免传统粒子群结合BP网络算法的"早熟"现象;通过与其他算法进行边坡稳定性评价的比较分析,表明了本文研究算法的可行性与合理性。

It is highly nonlinear and uncertain to evaluate and predict slope stability, and also difficult to express using accurate mathematical model. Firstly, the multiple slope engineering instances were adopted to constitute a learning sample set. The six main influence factors, including soil density, internal friction angle, cohesion, slope angle, slope height, void ratio, composed slope stability evaluation index. Then BP neural network model was optimized using particle swarm optimization algorithm to realize the hybrid algorithm. When maintaining the BP network algorithm of error back propagation correction weight, the network weights and threshold values were particles and updated using particle swarm algorithm global searching. At the same time the convergence speed was accelerated and the convergence precision was improved. The ＂premature＂ phenomenon for the BP network algorithm combining with traditional particle swarm was avoided. Finally, the feasibility and rationality of the proposed approach in the paper were verified in comparison with other slope stability evaluation algorithms.