应用改进的BP人工神经网络快速预测储层敏感性

标准的神经网络算法存在训练时间长，在一些特定的给定初值情况下会陷入局部最小等缺点，应用受到一定限制。该文介绍了改进的BP人工神经网络训练算法，该算法在一个权值修改过程中对权值修改两次，以达到加速收敛避免陷入局部最小等目的，利用该算法进行了储层敏感性快速预测软件研制。分析表明，该算法受人为因素干扰小，所需参数少，结果比较可靠，总体符合率达到91％，改进后的算法训练所需时间与标准BP网络相比缩短了许多，是一种适用于现场的良好方法。

Using improved BP artificial neural network in fast predicting the sensibility of reservoirs

The standard neural network takes long training time and will go into local minimum value in some special given initial value, it limits the use of the method. The training algorithm of improved BP artificial neural network is introduced in this paper, in which the weight value is amended twice in amended process of a weight value so as to accelerate the convergence and avoid going into local minimum value. And then the method is used to develop the fast prediction software of reservoir sensibility. The study shows that this method has little influence on subjective interference and requires lesser parameters; the result is more credible with 91% overall coincidence rate. The training time of improved algorithm is much shorter than that of standard method, it is a good method suitable for field.