Model Predictive Control-Based Coordinated Dynamic Optimization of Coal and Electricity Consumption for Low-Carbon Process Industry

As a typical high-emission process industry, cement manufacturing features strong non-linearity, time-varying delays, and nonstationary dynamics, with clinker calcination dominating energy costs and carbon emissions. This paper proposes a receding-horizon MPC coordinated optimization framework for reducing coal and electricity consumption during calcination. First, ReliefF screening extracts key energy-related state variables from high-dimensional DCS data under process-mechanism constraints. VMD and sliding windows are then used to construct multiscale dynamic features describing nonstationary and delayed process responses. An HHO-ELM model predicts future energy-consumption trajectories, while a GWO-based upper-layer optimizer searches for energy-efficient reference trajectories under operating constraints. A data-identified discrete-time plant-response model serves as the internal dynamic model of the lower-layer MPC, which calculates constrained control moves for tracking these references. The framework is evaluated using archived industrial DCS data and an offline receding-horizon closed-loop simulation. VMD-HHO-ELM yields the lowest prediction errors among the compared predictors, and GWO shows favorable convergence for economic reference generation. The simulated MPC results indicate coordinated reductions in coal and electricity consumption within the observed operating region.