Optimization of industrial waste composition for eco-cement production using machine learning with meta-heuristic method

Abstract

Industrial solid waste utilization in ecological cement production reduces carbon emissions and resource consumption. However, waste composition variability challenges product consistency, hindering widespread implementation. This study presents a novel artificial intelligence driven optimization framework that combines triple back-propagation neural network (BPNN), non-dominated sorting genetic algorithm-II (NSGA-II) and super-efficiency data envelopment analysis (SE-DEA), to optimize raw material proportions. The triple BPNN architecture, which separately models raw meal grinding, clinker calcination, and clinker homogenization, demonstrates superior prediction accuracy. By integrating SE-DEA within NSGA-II, the algorithm achieves balanced optimization across multiple performance metrics. Cement produced using this optimization framework demonstrates improved long-term mechanical properties, with 28-day compressive strength ranking in the top 4.11 % and 28-day flexural strength in the top 7.41 % of the historical production data from the same facility. The proposed framework exhibits significant potential for the development and promotion of ecological cement technology while promoting industrial waste utilization at large scale.