The　combustion　of　fossil　fuels　has　achieved　a　high　proportion　of　world＇s　electricity　generation,　and　different　types　of　atmospheric　pollutants　such　as　sulfur　dioxide　(SO2)　are　produced　during　this　process.　As　a　substantial　amount　of　SO2　is　released　by　coal　combustion,　flue　gas　desulfurization　(FGD)　technologies　are　extensively　used　in　coal-fueled　power　plants,　and　the　design　of　FGD　modeling　approach　plays　a　fundamental　role　in　optimization　and　control　of　the　desulfurization　process.　Nevertheless,　the　FGD　in　an　industrial　setting　is　a　rather　complicated　process　which　has　characteristics　of　non-linearity,　time　delays　and　non-stationarity,　rendering　dynamic　learning　of　an　FGD　process　a　truly　formidable　problem.　In　this　research,　a　novel　FGD　modeling　approach　integrating　partial　autocorrelation　function　(PACF),　gated　recurrent　unit　(GRU)　neural　network　and　attention　mechanism　(AM)　was　proposed　to　make　a　prediction　of　the　limestone　slurry　pH.　The　effectiveness　of　the　proposed　modeling　system　is　verified　through　the　investigation　of　a　real　FGD　process,　besides　the　structural　flexibility　and　outstanding　performance　of　our　method　demonstrates　its　widespread　application　prospects　in　different　industrial　scenarios.　　©　2024　IEEE.