Multi-objective optimization method for cement calcination system based on dual population differential evolution algorithm

The cement calcination system (CCS) demonstrates a high degree of coupling among operational indicators and experiences significant dynamic variations in its operating conditions. Traditional parameter‑setting methods based on empirical experience are insufficient for achieving coordinated optimization of energy consumption and product quality. To address these challenges, this study proposes a multi-objective optimization approach based on the Dual-Population Differential Evolution (DP-DE) algorithm, intended to ensure the CCS operates stably and efficiently in terms of energy consumption, while concurrently enhancing product quality. The proposed approach initially formulates a multi-objective optimization model that accounts for electricity consumption, coal consumption, and clinker quality, and integrates electricity and coal prices to weight the energy cost component. For the optimization process, a two-stage differential evolution algorithm employing a “decision-first, optimization-later” strategy is developed, in conjunction with a dynamic search-space partitioning mechanism to facilitate multi-step, smooth adjustments of controlled variable setpoints. To accommodate the nonlinear characteristics of complex industrial processes, Convolutional Neural Network(CNN) and Convolutional Neural Network-Long Short-Term Memory Network(CNN-LSTM)-based neural network fitness functions are constructed to capture relationships between process variables and target indicators from historical data, thereby enabling effective mappings from the solution space to the objective space. Experimental results indicate that, under stable operating conditions, this approach reduces energy costs by 3.1 % while maintaining clinker quality within acceptable limits. Furthermore, robustness experiments, which involve repeated trials with randomly initialized populations and minor input perturbations, confirm that the algorithm maintains consistent optimization trajectories and yields stable results under uncertainty, thereby demonstrating favorable engineering deployability.