基于多输出RBF神经网络的油藏参数反演方法OA
Reservoir parameter inversion method based on multi-output radial basis function neural network
为提高试井分析的反演精度和效率,提出一种基于多输出径向基函数(Radial Basis Function,RBF)神经网络的油藏参数反演方法.首先选定待反演的不确定参数及其范围,生成采样算例;然后利用梯度下降法训练出一个逼近真实函数关系的代理模型,即多输出 RBF 神经网络,构建代理模型预测值与实际观测值偏差的目标函数,再利用粒子群优化(Particle Swarm Optimization,PSO)算法搜索目标函数的最小值,得到一组较优参数值.与单输出 RBF 神经网络相比,多输出RBF 神经网络通过参数共享与同步输出完整压力响应序列,避免了为每个时间点压力单独建模的重复训练.对比实验表明,多输出 RBF 神经网络训练时间仅为单输出 RBF 神经网络训练时间的2%左右,且模型精度更高,预测曲线更平滑.通过油田现场实测数据验证可知,基于多输出 RBF 神经网络的油藏参数反演方法在复杂非均质油藏模型中能有效地对井底压力变化和压力导数进行拟合,为油藏参数智能反演提供了一种高效、可供参考的新方法,具有一定应用价值.
In order to improve the inversion accuracy and efficiency of well test analyis,this paper proposes a reservoir parameter inversion method based on a multi-output radial basis function(RBF)neural network.First,the uncertain parameters to be inverted and their ranges are selected to generate sampling cases.Then,a surrogate model that approximates the true functional relationship,i.e.,the multi-output RBF neural network model,is trained using gradient descent.An objective function is constructed based on the deviation between surrogate model predictions and actual observations,and particle swarm optimization(PSO)is employed to search for the minimum of the objective function to obtain a set of optimal parameter values.Compared with single-output RBF neural networks,the multi-output RBF neural network achieves parameter sharing and synchronous output of complete pressure response sequences,avoiding repetitive training required for modeling pressure at each time point separately.Comparative experiments demonstrate that the training time of the multi-output RBF neural network is only approximately 2%of that of the single-output RBF model,with higher model accuracy and smoother prediction curves.It is known from the verification with actual measured data from oilfied sites that the reservoir parameter inversion method based on the multi-output RBF neural network can effectively fit both bottomhole pressure variations and pressure derivatives in complex heterogeneous reservoir models.This method provides an efficient and practical new method for intelligent reservoir parameter inversion with significant application value.
沈婷;付宁;王凯;张康;李道伦
合肥工业大学数学学院,安徽 合肥 230601中国石油川庆钻探工程有限公司长庆井下技术作业公司,陕西 西安 710018中国石油川庆钻探工程有限公司长庆井下技术作业公司,陕西 西安 710018中国石油川庆钻探工程有限公司长庆井下技术作业公司,陕西 西安 710018合肥工业大学数学学院,安徽 合肥 230601
RBF神经网络多输出油藏参数反演历史拟合试井分析
RBF neural networkMulti-outputReservoir parameter inversionHistory matchingWell test analysis
《天然气与石油》 2026 (4)
70-81,12
国家自然科学基金"基于流动方程的致密油气藏深度学习数值模拟方法"(12372244)
评论