极限学习脊波过程神经网络及应用

Extreme Learning Ridgelet Process Neural Network and Application

为提高页岩的岩性识别精度,首先针对测井曲线连续变化、突变频繁的信号特征,利用脊波变换作为过程神经元的激励函数,提出一种脊波过程神经网络模型;其次通过Ada Boost的动态调整机制迭代调整模型和样本集权重,利用多个弱分类器的线性加权构建强分类器;最后为提高AdaBoost中的每个脊波过程神经网络模型的学习速度,提出一种基于满秩分解的极限学习算法,通过Moore-Penrose广义逆求解隐层输出权值。仿真实验以A区的B1井和B2井为例进行岩性识别,通过对比分析验证方法的有效性,识别效果优于其他过程神经网络模型,准确率最高可达90%左右。

To improve the accuracy about shale lithology identification,this paper proposes a ridgelet process neural network(RPNN)and selects the ridgelet transform as the active function of process neuron according to the signal characteristic of continuous and frequent variation about logging curve.RPNN uses the dynamic mechanism of AdaBoost to adjust iteratively weights of the models and the sample.In order to improve the RPNN learning speed in AdaBoost,an extreme learning algorithm based on full rank decomposition is proposed.The proposed method was applied to the lithology identification for B1 and B2 well in A area and its effectiveness was tested through comparison and analysis.The results show that the lithology identification result of RPNN is better than other process neural networks and the recognition accuracy of RPNN is up to about 90 per cent.