用约束型自组织特征映射网络实现多井约束插值

提出了一种用神经网络实现多井约束插值的新方法。作为传统的自组织特征映射网络(Self Organizing Featuere Map，SOFM)的一种推广形式，提出的约束型自组织特征映射(Constraint Self Organizing Featuere Map，CSOFM)对某些特定的输出节点及输出节点的排列方式加以约束，从而既保留了SOFM的长处(如自组织和拓扑映射)，又克服其在处理多井的约束插值问题时存在的一些缺点(如网络规模过大，训练时间过长和归位问题)。对实际资料的处理表明，其方法能取得满意的结果。

Application of Constraint Self-organizing Feature Map Neural Network in Multi-wells Interpolation

The paper proposes a novel Multi-interpolation method using neural network. As an extension of the traditional self-organizing feature map(SOFM), the proposed constraint self-organizing feature map(CSOFM) in the paper, with its partial output neurons and the output lattice constrained, keeps the good quality (such as self-oganizing and topographic map),and is seen to alleviate the shortcomings (such as too large net scale, too long training time and homing problem) of the former in the context of Multi-wells interpolation. The processing of real data shows this method can gain satisfactory result.