工場における排水処理では,微生物の代謝作用を利用 して回生用メタンガスを生成し,エネルギー消費量を抑えながら効率良く汚泥を除去する. しかし,大規模な工場では化学反応が複雑化し,処理設備の最適制御が困難となる.そのため,本研究では機械学習を用いて排水処理をモデル化し,制御において有効となる要因の抽出手法を検討している. 具体的には,汚泥除去率とエネルギー変換率を目的変数とし,排水処理の各工程での計測値を説明変数とした回帰モデルを構築する.このモデルにおいて,説明変数をフィルタリングしながら予測性能の変化を計測する事で,制御において有効な説明変数を抽出すると同時に,実際に制御を行った場合の目的変数に対する影響を模擬する手法の開発を進めている.

In the process of industrial wastewater treatment, using metabolism of the organism to remove the sludge is not only an efficient method, but the methane gas generated for regeneration also help to reduce the energy consumption. However, when the process scale becomes large, the chemical reaction becomes complicated, and the control process becomes more and more difficult to optimize. In this research, we are building a wastewater treatment model based on machine learning, and identifying the factors that affects the control efficiency. Specifically, while building a regression model, we set both the rate of removing sludge and the rate of energy conversion as the explanatory variables, and set the other measurements at each step of the wastewater treatment as the response variables. In this model, by measuring the changes in the prediction performance while filtering the explanatory variables, we extract the valid explanatory variables in control and simulate their impact on the target variables when performing the actual control.