基于改进的数量化理论和RBF神经网络组合方法的地下水水质预测

Predicting the trends of pollutant concentrations in groundwater based on the combined method of the improved quantification theory and RBF artificial neural network

文中首先运用了一种改进的数量化理论I模型作为预处理工具,对影响地下水水质的20个因子进行定性数据转换、数据降维,随后将8个重要特征因子作为RBF(径向基函数)神经网络模型的输入,进一步对监测井的采样数据进行学习、训练,揭示地下水污染质迁移转化规律。尝试用经过改进的数量化理论与RBF神经网络方法二者结合,对沈阳李官水源地研究区监测井地下水水质变化进行模拟与预测,其仿真结果覆盖了现有的绝大部分实测数据,适用范围广泛,具有一定的推广价值。

In this paper, an improved quantification theory I proposed by Chikio Hayashi was used as a preprocessing tool to covert quantitative data to qualitative data and to reduce data dimensionality for 20 factors impacting groundwater quality. Then 8 important characteristic factors were used as nodes of input layer in RBF Neural Networks, and RBF ANN model was created through training and learning the sampling data of monitoring well, finally migration and transformation law of pollutants were revealed. By using the combination model of the quantification theory and RBF ANN, the simulation results covered most of the existing experimental data and could forecast the dynamic changes of groundwater quality. The result is relatively accurate for a wide range and has some promotional value.