Construction of Expert System by means of Neural Networks for Repair Work and Multi-purpose Use of Deterioration Dam

研究了用神经网络方法构建劣化水库除险加固与综合利用专家系统的问题，首先通过问卷调查收集了人类专家对小型水库除险加固与综合利用的知识，采用ISM和FSM方法对知识进行了结构化，并对结构化后的知识进行了分类，最后对分类结果进行了分析。其知识的连接可以连接成前馈分层神经网络。采用神经网络的专家系统是通过学习数据集进行训练的，但由于BP（BackPropagation）算法训练时间过长，提出了一种新的训练算法--扩展的Kaiman滤波器（KNT）。结果表明，KNT具有较高的性能相比，BP。此外，专家系统集成的图像处理，统计分析功能和基础数据库的小水库，大坂市的Sakai作为一个例子。

The construction of an expart system by means of neural networks have been studied on repair works and multi-purpose use of deterioration dams.At first, knowledge of human experts on a small reservoir were collected by questionnaire surveys.Knowledge in regard to repair works and multi-purpose of small reservoirs were given structure using ISM and FSM method. The connection of its knowledge could be linked as feedforward layred neural networks. The expert system using neural networks were trained by learning data set.As training time, however, was taken much more in back propaqation (BP), a new training algorithm was developed by extended Kaiman filter (here called KNT). It was shown that the KNT had high performance compared with the BP.Further, the expert systems was integrated with an image processing, a function of statistical analysis and a fundamental data base for small reservoirs of Sakai City in Osaka as an example.