浄水プロセス制御を目的とした深層学習によるフロック画像からの凝集後濁度予測に関する研究

抄録

The objective of this paper is to predict the turbidity after flocculation from floc images in jar-test deep convolutional neural network (DCNN). Our goal is to develop a system to control the water purification process using the predictive model. In conventional studies, chemical parameters such as pH and alkalinity are generally used to predict turbidity. However, our proposed method does not use chemical parameters. It uses images of floating matter, called “floc”, generated during the flocculation process as input to the DCNN. We performed experiments using DCNN to predict turbidity after flocculation from images of “flocs” generated in jar-tests. We used VGG-16 as the DCNN for our experiments. Furthermore, we conducted experiments to compare the proposed method with the baseline method using chemical parameters.

As a result, 1) we revealed that the turbidity after flocculation can be predicted by using the image during flocculation as input to VGG-16; 2) we revealed that the optimal period to be used for the prediction model was the data 200 second after the start of the jar-tests; 3) we revealed that our proposed method can predict turbidity with better accuracy than the baseline method using chemical parameters.