改进的MVO-GRNN神经网络岩爆预测模型研究  

Research on improved MVO-GRNN neural network rockburst prediction model

作　　者：侯克鹏[1,2] 包广拓 孙华芬 HOU Kepeng;BAO Guangtuo;SUN Huafen(School of Land and Resources Engineering,Kunming University of Science and Technology,Kunming 650093,China;Yunnan Key Laboratory of Sino-German Blue Mining and Utilization of Special Underground Space,Kunming 650093,China)

机构地区：[1]昆明理工大学国土资源工程学院,昆明650093 [2]云南省中-德蓝色矿山与特殊地下空间开发利用重点实验室,昆明650093

出　　处：《安全与环境学报》2024年第3期923-932,共10页Journal of Safety and Environment

基　　金：云南省科技厅项目(KKS0202121020)。

摘　　要：准确预测岩爆烈度等级能有效指导岩爆灾害的防控。根据影响岩爆发生及烈度等级的3个因素构建岩爆评价指标体系,提出一种基于改进多元宇宙算法(Improved Multi-Verse Optimizer,IMVO)优化广义回归神经网络(General Regression Neural Network,GRNN)的岩爆预测模型。在普通多元宇宙算法(MVO)的基础上,运用自适应平衡机制调节MVO算法中的虫洞存在概率(V_(WEP))和旅行距离率(V_(TDR))两个重要参数来改进该算法;再运用改进的多元宇宙算法优化广义回归神经网络的光滑度,通过训练数据优选出最佳光滑因子σ,得到IMVO-GRNN神经网络岩爆烈度预测模型;最后结合工程实例验证模型的性能。研究表明,该模型相比传统模型寻优能力更强,精度更高,为岩爆预测提供了一种新的思路。The prevention and control of rockburst disasters are of great significance to the construction of deep underground engineering,and the accurate prediction of rockburst intensity level can effectively guide the prevention and control of rockburst disasters.Based on the comprehensive consideration of the occurrence mechanism and the judgment basis of rockburst,the evaluation index system of rockburst intensity grade was constructed with three main factors:the stress coefficient of rock(σ_(θ)/σ_(c)),the brittleness coefficient of rock(σ_(c)/σ_(t))and the index of rock elastic energy(W_(et)).These factors affect the occurrence and intensity class of rock bursts.A rockburst prediction model based on an improved Multi-Verse Optimizer algorithm(IMVO)optimized Generalized Regression Neural Network(GRNN)is proposed.Firstly,122 groups of existing rockburst cases were collected from domestic and overseas literature as the sample data of the model;Then,the adaptive balancing mechanism was used to adjust the wormhole existence probability(V_(WEP))and travel distance rate(V_(TDR))in the Multi-Verse Optimizer(MVO)to improve the algorithm and it also improves the global optimization ability of traditional Multi-Verse Optimizer.To make the network layer of the GRNN good information transmission and better mining data information,the improved Multi-Verse Optimizer was used to optimize the smoothness of activation function between neurons in the generalized regression neural network.Besides,the sample data were divided into two data sets for training the prediction model and cross-validation,the optimal smooth factorσof the prediction model was selected,and the prediction model of rockburst intensity grade based on IMVO-GRNN neural network was obtained.Finally,the feasibility and generalization performance of the model is verified by two practical cases with different engineering backgrounds.The results of the research show that the model has greater optimization ability and higher prediction accuracy than the traditional mo

关 键 词：安全工程 岩爆预测 多元宇宙算法 广义回归神经网络(GRNN) 虫洞存在概率 旅行距离率 

分 类 号：X45[环境科学与工程—灾害防治]