时间推移试井参数预测

Prediction of Parameters from Time,Lapse Well Test.

针对压力监测井点的动态监测信息在时间上的不连续性和历史的继承性,首次提出了“时间推移试井”理论,并把神经网络技术应用到试井分析理论中,建立了“时间推移试井”的BP反馈神经网络结构模型。把试井的历史信息作为学习的样本,给出了优化的网络结构。结合模拟退火算法,避免了局部最优。在油田开发过程中,利用“时间推移试井”分析方法,可以对动态参数在时间上的变化进行推移预测。通过实例分析,实现了由单次试井分析向多次试井分析的拓展。

Considering the information discontinuity and historical inheritance of pressure monitoring performance,this paper firstly proposes the theory of 'time,lapse well test (TLWT)',and introduces neural networks into well test analysis,hence,has developed a model of BP feedback neural network structure for TLWT. Taking the historical test information as the sample of learning,the optimal network structure is suggested. By combining with simulated annealing approach,local optimization can be avoided. During an oilfield development,utilizing the method of TLWT,time,lapse prediction of performance parameters can also be made. The case analysis shows that the extrapolation from the individual to the multiple analysis of well test has been realized.