Abstract
The proper treatment of wastewater has emerged as a significant concern across numerous industries, particularly semiconductor manufacturing, which generates a substantial amount of wastewater during wafer cleaning processes. The efficiency of wastewater treatment can be negatively impacted by highly concentration of hydrogen peroxide which poses notable environmental hazards. Accurately predicting the concentration of hydrogen peroxide is crucial for making informed timely decisions regarding appropriate treatment. Time Series Data has been utilized in previous studies along with AI models for precise prediction. However, the selection of appropriate parameters within these AI models remains a challeng, often regarded as a black box. Accordingly, the UNISON framework, which integrates the modelbased multi-object particle swarm optimization (MOPSO) algorithm with long short-term memory, is designed to select the optimized lags and parameter for more reliable models. Furthermore, the training time or latency are considered in this research to sever the smart manufacturing systems. A comprehensive empirical study was undertaken at an industryleading semiconductor company to determine the validity and effectiveness of the proposed model. The results demonstrated that the proposed model achieved higher accuracy in predictions and faster training times, even with half the number of parameters tuned. The proposed model has proved its practical viability and demonstrated robust performance for better wastewater treatment and sustainability.