摘要
瓦斯含量预测取决于多因素、非线性的函数关系的建立 ,预测模型建立的准确与否决定于各个影响因素之间的相互作用、相互耦合的特性。文中将神经网络与遗传算法有机地结合起来 ,以神经网络理论为基础 ,利用遗传算法优化隐含层神经元个数和网络中的连接权值 ,建立瓦斯含量预测模型。在实验室测试数据的基础上 ,建立遗传神经网络训练和检验样本集 ,其中包含有 38个典型样本 ,并且将检验结果分别与回归模型、标准BP神经网络、自适应BP神经网络的预测结果进行比较。结果表明 :遗传神经网络模型可靠 ,预测精度高 ,为促进软计算技术与瓦斯地质的结合奠定了基础。
Forecasting the gas content depends on an establishment of a non liner functional relation of many factors; the accuracy of the forecasting model for the gas content is determined by the peculiarities of the interaction and coupling between all the affecting factors. This paper combines neural networks with genetic algorithm; on the basis of NN theory, and applying GA to optimize the construction and the power size of NN, a forecasting model of gas content is established. On the basis of the data in laboratory, the training and testing samples of GA NN model are founded, including 38 typical specimens. The verifying outcome has been compared with the output of back assay model, normal BP NN, and auto adapting NN. The result shows that the GA NN model is reliable and precise, which founds the basis for promoting the integration of soft calculation and gas geology.
出处
《地学前缘》
EI
CAS
CSCD
2003年第1期219-224,共6页
Earth Science Frontiers
