Upgrading processes to reduce biodegradable organic substance addition is crucial for treating landfill leachate with high pollutant concentrations, aiding carbon emission reduction. Aggregate size in activated sludge processes impacts pollutant removal and sludge/water separation. This study investigated microbial community succession and driving mechanisms in different floc-size aggregates during nitrogen removal progress upgrade from conventional to partial nitrification–denitrification in a full-scale landfill leachate treatment plant (LLTP) using 16S rRNA gene sequencing. The upgrade and floc sizes significantly influenced microbial diversity and composition. After upgrading, ammonia-oxidizing bacteria were enriched while nitrite-oxidizing bacteria suppressed in small flocs with homogeneity and high mass transfer efficiency. Larger flocs enriched Defluviicoccus, Thauera, and Truepera, while smaller flocs enriched Nitrosomonas, suggesting their potential as biomarkers. Multi-network analyses revealed microbial interactions. A deep learning model with convolutional neural networks predicted nitrogen removal efficiency. These findings guide optimizing LLTP processes and understanding microbial community dynamics based on floc size.