同伦BP算法在储集层参数预测中的应用

Application of Homotopy BP Algorithm to Prediction of Reservoir Parameters

同伦算法是解决非线性优化问题的新理论,将同伦算法引入误差反向传播神经网络,提高了网络的训练速度,消除了常规BP网络算法收敛于局部极值的缺点。该算法具有稳定性强、训练速度快、收敛精度高的优点。实际应用表明,利用此网络进行储集层参数预测,提高了参数预测的可靠性;该算法具有克服网络病态的能力。

Homotopy algorithm is a new method for solving nonlinear optimization problems. This paper introduces the homotopy algorithm into back-propagation neural network and indicates that the convergence speed can be improved and the defects of converging to local extremum can be eliminated. It is of rapid convergence speed, high convergence precision and good stability. Application of the algorithm to predicting reservoir parameters could improve the reliability of the parameter prediction. Case study proves its effectivity in prediction of reservoir parameters.