论文部分内容阅读
为提高总有机碳含量(TOC)的预测精度,针对测井曲线的时变、奇异性特征,选用脊波函数作为过程神经元的激励函数,提出一种连续脊波过程神经元网络.模型训练方面首先给出基于正交基展开的梯度下降法;其次为提高模型训练收敛能力,提出一种沿Bloch球面纬线实施莱维飞行的量子衍生布谷鸟算法,并用于模型参数优化;最后将训练好的脊波过程神经网络应用于泥页岩TOC预测,通过相关性选取对TOC响应敏感的测井曲线作为模型特征输入.实验对比结果表明,该方法的预测精度较高,较其他过程神经网络提高7个百分点.
In order to improve the prediction accuracy of total organic carbon (TOC), according to the time-varying and singularity characteristics of logging curves, a ridge function is chosen as excitation function of process neurons, and a neural network of continuous ridge processes is proposed. Firstly, the gradient descent method based on orthonormal basis expansion is given. Secondly, in order to improve the ability of model training convergence, a quantum derivative cuckoo algorithm based on Bloch spherical latitude for Levi flight is proposed and used to optimize the model parameters. Finally, The ridge-wave process neural network is applied to the TOC prediction of shale, and the log curves that are sensitive to TOC response are selected as the model features by the correlation.Experimental comparison results show that the method has higher prediction accuracy than the other process neural networks 7 percentage points.