数据缺失的SOA-KELM岩爆倾向性预测

:为简化模型结构、解决迭代训练拖延问题，利用海鸥（SOA）算法进行核极限学习机（KELM）重要参数择优，建立基于数据插补和SOA-KELM的岩爆风险预测模型。综合岩爆预测过程中多因素影响，选取单轴抗压强度，单轴抗拉强度等6种指标作为岩爆风险评价指标，搜集93组岩爆实测样本。一方面采用随机过采样补充少数类别样本数据，一方面采用ELMAN神经网络进行缺失数据插补，构建高质量岩爆风险预测样本数据库。最终将预处理后的数据输入4种模型中进行分类预测。结果表明：数据插补后，各模型预测准确率提升5.56%~16.67%。不同情况下，SOA-KELM预测准确率均为最高数值，且数据随机过采样处理并未影响模型预测准确率，融合ELMAN神经网络和SOA-KELM的预测模型可有效应用于岩爆风险预测，为实际岩爆预测提供了新思路。

SOA-KELM Rockburst Risk Prediction with Missing Data

:In order to simplify the model structure and solve the problem of iterative training delay, the seagull (SOA) algorithm is used for nuclear extreme learning machine (KELM) important parameter selection, and a rockburst risk prediction model based on data interpolation and SOA-KELM is established. Comprehensive rockburst prediction process of multi-factor influence, selected uniaxial compressive strength, uniaxial tensile strength and other six indicators as rockburst risk evaluation index, the collection of 93 groups of rockburst actual measurement samples. On the one hand, random oversampling is used to supplement the data of a few categories of samples, and on the other hand, ELMAN neural network is used to interpolate the missing data to build a high-quality rockburst risk prediction sample database. Finally, the pre-processed data were input into four models for classification prediction. The results show that the prediction accuracy of each model is improved by 5.56%~16.67% after data interpolation. The SOA-KELM prediction accuracy was the highest value in different cases, and the random data oversampling process did not affect the model prediction accuracy, and the prediction model integrating ELMAN neural network and SOA-KELM can be effectively used for rockburst risk prediction, which provides a new idea for practical rockburst prediction.