Enhancing Boiler Operational Economy via Data-Driven Collaborative Combustion Optimization

The growing need for flexible operation and reliable steam supply challenges coal-fired boilers to maintain an operating economy through fixed or single-variable combustion adjustments. This study develops a condition-aware, data-driven framework for coordinated combustion optimization. Historical stable-operation samples from the boiler were first classified according to the main steam flow rate and apparent fuel-related operating state. For each operating category, an MSE-constrained RF-NGBoost surrogate model was developed to predict the coal fan variable operating cost. The trained surrogate models then served as the offline response environment for a twin delayed deep deterministic policy gradient (TD3) agent, which searched for coordinated target set points of flue-gas oxygen content, bed temperature, and bed pressure drop under operating constraints. The RF-NGBoost models achieved R ² values of 0.967-0.982, RMSE values of 0.387-0.483 CNY/t, and mean absolute error values of 0.255-0.396 CNY/t across the six operating categories. For representative low-, medium-, and high-load conditions, the offline TD3 optimization identified condition-specific target set points associated with predicted coal-electricity operating cost reductions of 0.36-1.07% within the historical feasible domain. In an on-site comparison comprising two representative two-day operating periods, the TD3-derived target-set point strategy was associated with a 0.51% reduction in the actual coal-electricity operating cost compared with manual experience-based adjustment while maintaining main-steam supply and predefined operational constraints. The proposed framework links condition-specific cost modeling, offline coordinated optimization, and field verification to support economically oriented combustion adjustment in circulating fluidized bed boilers.