BP神经网络法预测水基绒囊钻井液当量静态密度

为研究水基绒囊钻井液当量静态密度随井深变化规律，以绒囊钻井液PVT实验数据为样本，建立了反映绒囊钻井液压力、温度和密度关系的BP神经网络。以BP神经网络为基础，建立与压力和温度相关的井深与绒囊钻井液井下静态当量密度预测模型。模型计算结果相比多元回归预测结果，与PVT实测数据相对误差更小，与磨80-C1井现场测定的0~2 500 m钻井液静态当量密度结果更吻合。用建立的BP神经网络模型预测磨80-C1井所在的磨溪地区绒囊钻井液2 500~6 000 m的静态当量密度，发现绒囊钻井液随井深增加密度逐渐减小，表明绒囊未在高温高压下被压缩成连续相，随井深增加，温度使气囊膨胀作用比压力压缩作用更明显，间接证明了绒囊结构抗压缩能力强的同时，也表明温度加强了绒囊封堵作用。同时，BP神经网络应用于预测钻井液井下静态当量密度，为井下密度预测提供了一种新的数学处理方法。

Prediction of equivalent static density of water-based fuzzy ball drilling fluid by BP neural network method

In order to study the equivalent static density changing rule of water-based fuzzy ball drilling fluid with different well depth, BP neural network is established, based on the PVT experimental data of fuzzy ball drilling fluid, reflecting the relationship between pressure, temperature and drilling fluid density. On the basis of the network，the equivalent static density prediction model in different well depth is established, and the well depth is associated with pressure and temperature. Compared with the result predicted by multiple regressions, the BP neural network method can obtain higher accuracy, and are more consistent with the static equivalent density in Well M80-C1 at the depth of 0~2 500 m. We predicted the equivalent static density of water-based fuzzy ball drilling fluid in Moxi area at the depth of 2 500~6 000 m by BP neural network, discovering it decreased with depth increase, which proved the fuzzy ball was not compressed into continuous phase at HTHP. The gas ball expansion capacity by temperature was greater than that by pressure with well depth increase, which proved the fuzzy ball structure resistance to compression is strong and temperature can strengthen the blocking effect of fuzzy ball. BP network provides a new mathematical method to predict the downhole equivalent static density of water-based fuzzy ball drilling fluid.