基于HF-RBF的初始地应力场预测研究与应用

用水压致裂法(HF)测量地应力时存在实测数据数量少、不连续、精度低的问题.为此,建立基于HF-RBF的初始地应力场预测方法,首先构建埋深和水平主应力之间的学习样本,然后利用径向基函数(RBF)神经网络对学习样本进行训练,建立埋深与水平主应力之间的非线性映射,最后利用检验样本对初始地应力场预测效果进行检验.在黑龙江某矿山利用2个勘探钻孔开展了水压致裂法测量地应力试验,应用HF-RBF方法获得了该矿区水平主应力分布规律,训练样本最大预测误差为28.6%,检验样本水平最小、最大主应力预测误差都在10%以内,满足工程设计要求.表明了在矿山钻孔勘探期间,HF-RBF法能较好地预测矿区初始地应力场.

Research on the Prediction of Initial Stress FieldBased on HF-RBF and Application

There remains a problem that the discreteness of measured data is high and the calculation accuracy of initial stress is low when it is tested using hydraulic fracturing(HF) technique.Therefore,a method,that first a learning sample between buried depth and horizontal principal stress is set up,then it is trained by using RBF Neural Network,finally effect of predicting initial stress field is tested with the test sample,is established to predict initial stress field based on HF-RBF.With two exploration boreholes of a certain mine of Heilongjiang province,initial stress test by hydraulic fracturing technique is conducted and through the HF-RBF method,the distribution of horizontal principal stress of the mine area is obtained,which reveals that the maximal prediction error of horizontal maximum principal stress and horizontal minimum principal stress is 27.8% and 22.8% and the result can meet the requirement of engineering design.The above showa that HF-RBF method can get a good forecast of mine initial stress field during mine exploration and drilling.