| Abstract |
| In this study, the concentration of dissolved oxygen (DO) at different water depths in a small river connected to the Nakdong River was monitored with parallel jet streamer device to improve water quality. DO probes were installed in correspondence of the upper, middle, and lower sections of the river at the different depths and operated for 2 months. To determine the stagnation of water in the river, we produced DO graphs for different depth intervals. Overall, we prepared 343 graphs, identifying 7 intervals with characteristic dissolved oxygen concentrations, including a stagnant zone. We separately applied an artificial neural network (ANN) and a convolution neural network as learning models: in the first case, a correct answer rate of only 29.2% was obtained from the derived weight and bias, while in the second case it corresponded to 94.5%. The learning graphs were randomly selected from 40 to 300. The correct answer ratios were 94.8%, 91.3%, and 88.6% for 250, 200, and 50 graphs, respectively. By applying the control logic to the actual monitoring results, we decided to label as a “stagnant region” the depth interval characterized by correct answer ratios comprised between 84.9% and 83.5% (i.e., depths between 30 m and 60 m). |
|
|
| Key Words |
| ANN, CNN, Disolved Oxygen, Monitoring, Stagnation, 인공신경망, 합성곱층 신경망, 용존산소, 모니터링, 정체수역 |
|
|
|
|
|
PDF