Process Data Correlation Analysis for Semi-Supervised Modeling of Dioxin Emissions in MSWI Processes

The pollution caused by municipal solid waste (MSW) has become a global issue that requires urgent attention and resolution. MSW incineration (MSWI) technology has become the preferred method due to its advantages of high mass reduction, high volume reduction, and energy recovery. The trace pollutant dioxin (DXN) emitted from this process is often referred to as the “poison of the century”. However, due to limitations in detection technology and the high costs associated with detection, challenges arise in modeling, such as the scarcity of samples, high-dimensional features, and the “memory effect”. How to effectively utilize the vast amount of unlabeled process data remains an area that has yet to be fully explored. To address these issues, this article proposes a process data correlation analysis method for semi-supervised modeling of DXN emissions. First, the original process data is divided into positive and negative time-scale segments, with the division centered around process data that contains real markers. Second, feature selection based on correlation and mutual information (MI) is performed on each group of positive and negative time-scale data after the division. Finally, a smallsample data-driven DXN model is constructed, and a correlation evaluation is performed based on its prediction performance. Validation based on historical data from an MSWI plant in Beijing shows that the relative fitting error between the negative time-scale data and the true value data is smaller in the testing set.