基于组合赋权的混合粒子群优化支持向量机的岩爆倾向性预测

为有效预测岩爆灾害发生烈度,提出一种基于组合赋权的混合粒子群优化支持向量机(H-PSO-SVM)岩爆倾向性预测模型。根据岩爆发生机制,在分析岩爆发生的主要影响因素的基础上确定出评判指标;综合考虑模糊层次分析法(FAHP)所得主观权重和熵权法所得客观权重,应用调和平均数概念,构建组合赋权准则;引入遗传算法交叉、变异操作改进传统粒子群(PSO)极值跟踪和粒子更新方法,建立H-PSO-SVM岩爆倾向性预测模型。利用国内外已有工程实例数据进行50次随机抽样试验,对比分析H-PSO-SVM模型和PSO-SVM模型等预测结果。结果表明:H-PSO-SVM模型应用于岩爆工程实例预测具有可行性和适应性,模型预测的准确率高于其他模型,且预测结果更稳定。

Forecast research on the rock burst liability based on the comprehensive evaluation H-PSO-SVM Model

This paper aims to propose a rock-burst liability prediction model in association with the Combination Weighing factors based on the H-PSO-SVM theory so as to predict and forecast the intensity of the rock-burst effectively. For the said purpose, we have first of all to analyze the main factors leading to the internal and external eruption forms from the point of view of the mechanism leading to the rock-burst. And,then,it is necessary for us to determine the Stress Coefficient (σθ /σc) further,so as to identify and determine the intensity of the rock-burst Fragility Coefficient (σc /σt) and the Wall Rock Impact Tendency Index (Wet). For it would only be made possible to construct the Combination Weighting rule by means of the concept of Harmonic Mean based on the subjective weight by Fuzzy Analytic Hierarchy Process (FAHP) and the objective weight by the Entropy Weighing method. Furthermore,the paper has mainly introduced the cross and mutation operations of the Genetic Algorithm (GA) to improve the extreme value tracking and the particle update methods of the traditional Particle Swarm Optimization (PSO) so as to boost the optimal performance of the parameters of c,g in the model SVM and simultaneously establish the H-PSO-SVM Rockburst prediction model. What is more,the paper has also carefully observed and examined the stability of the accuracy rate of the results of the experiences gained in the different times,mainly by using the recurrence estimation and the predictive estimation based on the standard deviation of retrogressive and progressive detection and forecast accuracy. Thus,finally,it has been made possible for us to determine and confirm that the optimum results can be gained through 50 times of sampling experiments. Furthermore, according to the available projection data gained both from home and abroad,we have conducted 50 random experiment samples to compare and analyze the prediction results of the H-PSOSVM model,the SVM model,the PSO-SVM model,the GA-SVM model and the Grid-SVM model. The results of the analysis of the above samples show that: (1) It is available and adaptable for the H-PSO-SVM model to be used to predict the rock-burst intensity. (2) The retrogressive detection accuracy (97. 277 8％) and the prediction accuracy (83. 800 0％) of H-PSO-SVM model both prove to be remarkably more advantageous than the results gained by using all the other models,which may account for more accuracy of the predictive results gained by using H-PSO-SVM model than all the other ones. (3) The retrospective detection accuracy standard deviation (0. 031 7) proves to be obviously lower than that (0. 109 3) gained by other models,which proves that the predictive results of H-PSO-SVM model are more stable and regular than those gained by using other ones.