Predicting Marine Water Quality in Taichung Harbor Using Propagation Neural Network

• Main Authors: Chun-Yi Li, 李俊毅
• Other Authors: Tzu-Ti Pai
• Format: Others
• Language: zh-TW
• Online Access: http://ndltd.ncl.edu.tw/handle/92984226700870763723

碩士 === 朝陽科技大學 === 環境工程與管理系碩士班 === 98 === Neural network is one kind of relation mode that can quickly and accurately reflect input-output, and mainly operates through the training, constantly adjusts the value (weight and leaning among joint), making the exportation more approach target exportation that the network calculates. This research establishes estimate mode of pH and dissolved oxygen (DO) value for the marine water quality in Taichung harbor to predict local fluid variety by applying Back-Propagation Neural Network (BNN). The research shows that the best mean absolute percentage error (MAPE) 1.46% and Pearson’s correlation (R) 0.34 of training by using four input parameters to compare one output parameter in pH of west pier. In predict part, the best result will be MAPE 0.15% and R 0.53 by comparing three input parameters and one output parameter in main ship canal. For DO figure, the best of training result will be MAPE 7.39% and R 0.39 by comparing two input parameters and one output parameter in exit of fishing port. Moreover, the best result of prediction will be MAPE 3.79% and R 0.73 by comparing five input parameters and one output parameter in west pier. According to training and prediction of pH and DO figure, the simulation result is very close to practical monitor. Therefore, the mode makes a conclusion that predicts effect perfects.