Combination of Kohonen network model and Back-propagation neural network model to predict Biochemical oxygen demand(BOD5) removal rate of sewage resource from the horizontal subsurface flow(HSSF) constructed wetlands

This study used Kohonen network model Self-Organizing Feature map(SOFM) with Principal Components Analysis(PCA) and Back-Propagation Network(BPN) to predict Biochemical oxygen demand(BOD₅) removal rate of water resource from the surface-flow constructed treatment wetlands at Yuli Township in Hualien County from December, 2005 to June, 2006. With this self-organizing map technology, the water quality parameters were lowered from 7-D to 2-D and during which all parameters were automatic clustered. The principal components analysis(PCA) was selected to analyze selected parameters from above typical 2-D cluster parameters. After the max variable from the above selected parameters was determine it was incorporated into a layer of BPN to continue varied processes. In short, in order to predict nitrified effluent BOD₅ removal rate precisely, all of above measurable variables need to under go the steps of (SOFM-PCA-BPN) sequentially.