【摘要】：為提高總有機(jī)碳含量(TOC)的預(yù)測精度,針對測井曲線的時變、奇異性特征,選用脊波函數(shù)作為過程神經(jīng)元的激勵函數(shù),提出一種連續(xù)脊波過程神經(jīng)元網(wǎng)絡(luò).模型訓(xùn)練方面首先給出基于正交基展開的梯度下降法;其次為提高模型訓(xùn)練收斂能力,提出一種沿Bloch球面緯線實施萊維飛行的量子衍生布谷鳥算法,并用于模型參數(shù)優(yōu)化;最后將訓(xùn)練好的脊波過程神經(jīng)網(wǎng)絡(luò)應(yīng)用于泥頁巖TOC預(yù)測,通過相關(guān)性選取對TOC響應(yīng)敏感的測井曲線作為模型特征輸入.實驗對比結(jié)果表明,該方法的預(yù)測精度較高,較其他過程神經(jīng)網(wǎng)絡(luò)提高7個百分點(diǎn).
[Abstract]:In order to improve the prediction accuracy of total organic carbon content (TOC), a continuous ridgelet process neural network is proposed in this paper, which is based on the time-varying and singular characteristics of log curves, and the ridgelet function is selected as the excitation function of the process neurons. In the aspect of model training, the gradient descent method based on orthogonal basis expansion is presented firstly, and then, in order to improve the convergence ability of model training, a quantum derived cuckoo algorithm is proposed to perform Levi flight along the Bloch spherical weft, and it is used to optimize the model parameters. Finally, the trained ridgelet process neural network is applied to shale TOC prediction, and logging curves sensitive to TOC response are selected as model feature input by correlation. The experimental results show that the prediction accuracy of this method is higher than that of other process neural networks by 7 percentage points.
【作者單位】： 東北石油大學(xué)計算機(jī)與信息技術(shù)學(xué)院;山東科技大學(xué)信息科學(xué)與工程學(xué)院;中國石油大學(xué)(華東)非常規(guī)油氣與新能源研究院;
【基金】：國家自然科學(xué)基金項目(61170132,41330313) 黑龍江省自然科學(xué)基金項目(F2015021)
【分類號】：P631.81;TP183
，

本文編號：2223937