BP人工神经网络在富水性评价中的应用

Application of back propagation artificial neural network in water abundance evaluation

为了对研究区含水层进行富水性评价,并减轻评价结果对水文孔的依赖程度,通过对地质及水文地质资料的分析,确定出四个富水性主控因素,分别为含水层厚度、岩芯采取率、脆性岩厚度比和风化影响指数。引入具有自主学习、非线性映射能力的BP人工神经网络,将25组经量化、归一化处理的主控因素数据作为网络学习样本、以实测单位涌水量为预测目标,通过反复训练学习,实现了对主控因素到单位涌水量映射关系的精确模拟。最后,使用训练好的神经网络对研究区富水性进行了预测,并引入灵敏度分析方法分析了预测结果对主控因素的敏感性。

In order to evaluate the water abundance of the study area,and to reduce the dependence of the evaluation results on the hydrographic bore. After making the best of geological and hydrogeological data,four main controlling factors is established,which are aquifer thickness, core recovery percentage, thickness ratio of brittle rock and weathering influence index. BP artificial neural network with autonomous learning,nonlinear mapping ability is introduced into the study,the data of 25 groups of main control factors treated by quantification,normalization is used as a network of learning samples,taking the measured specific well yield as the prediction target, the simulation of the mapping relationship between the main control factor and the specific well yield has been realized by repeated training. Finally,the trained neural network is used to predict the water abundance,and the sensitivity analysis method is introduced to analyze the sensitivity of the prediction results to the dominant factors.