改进的GA-BP神经网络在矿井突水水源判别中的应用

矿井突水水源的判别是制定防治水措施的重要环节。通过对某矿含水层水化学特性的相关性分析，将PCA算法、K折交叉验证算法嵌入GA-BP神经网络，提出了一种新的GA-BP神经网络，将其应用于实例分析中，并与传统的方法进行比较。结果表明:针对水化学特性相近的含水层，PCA算法能够排除样本中的冗余信息，降低样本指标维度，简化BP神经网络结构；K折交叉验证算法能够提高GA算法对BP神经网络权值的寻优质量，使GA算法的进化方向更具合理性；二者的引入大大优化了传统GA-BP神经网络性能，其判别精度更高、适用性更强、结果更可靠，在矿井突水水源判别方面具有很好的应用前景。

Application of improved GA-BP neural network on identification of water inrush source in mine

The identification of water inrush source in mine is an important link in formulation of water prevention and control measures. Through the correlation analysis of hydrochemical characteristics of the aquifer in a mine, the PCA algorithm, k-fold cross validation algorithm were embedded into the GA-BP neural network. A new GA-BP neural network was proposed and applied to an example analysis, then compared with the traditional methods. The results showed that for the aquifers with similar hydrochemical characteristics, the PCA algorithm can eliminate the redundant information from the samples, reduce the dimension of sample index, and simplify the structure of BP neural network. The k-fold cross validation algorithm can improve the optimization quality of GA algorithm for weights of BP neural network, and make the evolution direction of GA algorithm more reasonable. The introduction of both the algorithms greatly optimize the performance of traditional GA-BP neural network. The method has higher identification accuracy, stronger applicability and more reliable results, and it has a good application prospect for water inrush source identification in mine.