基于遗传BP神经网络模型的矿井突水水源判别

Discriminating Mine Water Inrush Source Based on Genetic BP Neural Network Model

在分析矿井突水水源的水化学特征基础上,选取常用的Na^++K^+,Ca^(2+),Mg^(2+),Cl^-,HCO_3^-,SO_4^(2-)等6种离子质量浓度作为水源判别的依据,将具有局部搜索能力的BP神经网络和具有全局寻优功能的遗传算法(GA)进行结合,提高神经网络的泛化性。为了验证其优点,分别采用BP和GA-BP两种神经网络模型对20组训练样本进行训练,并对6组待测样本进行判别。结果表明:GA-BP神经网络模型克服了BP神经网络初始权值与阈值的随机性、易陷入局部最优的缺点,能提高BP神经网络的判别精度;虽然经过GA初始化的BP神经网络在训练过程中收敛速度与误差均不如未优化的BP神经网络,但GA-BP网络模型泛化性却高于BP网络模型,能提高突水水源的判别准确性。

Based on analysis of the water chemical characteristics of mine water inrush, the commonly used six kinds of ion concentrations, Na＋ K＋ ,Ca2＋ ,Mg2＋ ,Cl- ,HCO3 - and SO42- , are selected as the basis of mine water dis- crimination, and the BP neural network having the local searching ability is combined with the genetic algorithm （GA） having the global optimization function, to improve the generalization of neural network. In order to verify its advantages, 20 groups of training samples are trained respectively by BP, GA - BP neural network model, and 6 groups of test samples are tested. The results show that, GA - BP neural network model can overcome the defects, the randomness of BP neural network initial weights and thresholds and being easy to fall into the local optimum, and can improve the precision of the BP neural network. Although the convergence speed and error of BP neural network which is initialized by GA initialization is not so good as BP neural network which is not optimized in the training process, the generalization of GA - BP network model is higher than that of BP network model, and so it can improve the accuracy of water inrush source.