针对排土场边坡稳定性分析，提出了一种利用主成分分析法降低数据冗余性、粒子群算法优化极限学习机权值阈值的PCA-PSO-ELM排土场边坡稳定性预测模型。确定了土壤黏聚力、内摩擦角、排土场斜角、地基承载力、地震烈度、降雨和降雪条件、排土工艺以及乱采乱挖状况8个排土场稳定性预测指标，针对100组相应排土场数据，采用训练时间、RMSE值和决定系数R ²来评价和对比PCA-PSO-ELM模型与BP神经网络模型、ELM模型和PSO-ELM模型预测结果的有效性。研究结果表明：利用经PCA降维处理过的排土场稳定性样本数据作为输入变量去训练和测试PSO-ELM网络模型，预测值与真实值非常接近，其预测精度和效率不仅高于ELM算法，而且远远优于传统BP神经网络算法。经过PCA法优化的PSO-ELM模型与未经PCA处理过的PSO-ELM模型相比，前者在效率相差甚微的基础上大幅缩短了计算时间，证明了该方法具有一定的实用价值。

Aiming at the stability analysis of dump slope，a PCA-PSO-ELM dump slope stability prediction model is proposed in this paper，which uses principal component analysis method to reduce data redundancy and particle swarm optimization algorithm to optimize the weight threshold of extreme learning machine. Eight prediction indexes of dump stability were determined in this model，including soil cohesion，internal friction angle，dump slope angle，foundation bearing capacity，seismic intensity，rainfall and snowfall conditions，dumping technology and random mining and digging conditions.According to 100 groups of corresponding dump data，training time，RMSE value and determination coefficient R ² were used to evaluate and compare the validity of prediction results of PCA-PSO-ELM model，BP neural network model，ELM model and PSO-ELM model.The research results show that as the input variable to train and test the PSO-ELM network model，the dump stability sample data processed by PCA dimensionality reduction，made predicted value very close to the real value.The prediction accuracy and efficiency are not only higher than the ELM algorithm，but also far better than the traditional BP neural network algorithm.Compared with the PSO-ELM model without PCA treatment，the PSO-ELM model optimized by PCA method can significantly shorten the calculation time on the basis of little difference in efficiency，which proves that the method has certain practical value.

http://www.goldsci.ac.cn/article/2021/1005-2518/1005-2518-2021-29-5-658.shtml