Coal-fired power remains a major source of electricity in China, making accurate NOx prediction essential for combustion tuning and environmental compliance. Leveraging an Industry 4.0 data environment with plant Distributed Control System (DCS) measurements, this study develops an intelligent prediction framework for boiler-outlet NOx. After 3σ-based outlier handling and Min–Max normalization, we estimate variable time delays within a 15-minute window using the Maximum Information Coefficient (MIC) to capture dynamic correlations. We then propose a Time-Varying Causal Feature Decoupling Network (TV-CFDN) that constructs time-conditioned causal graphs and extracts dominant spatiotemporal modes (via multi-window dynamic mode decomposition), yielding interpretable embeddings and a ranked set of 40 key features. Building on these representations, an Attention-Integrated LSTM (ALSTM) jointly predicts boiler-outlet NOx and main steam temperature, enhancing sensitivity to critical variables and periods. Experiments on real plant data demonstrate superior performance over LSTM, DBN, DNN, and ELM baselines, confirming the efficacy of time-delay reconstruction and causal feature decoupling. The results provide intelligent, data-driven predictive decision support for combustion optimization and regulatory compliance by offering interpretable feature importance rankings and dynamically attending to critical operational variables and time periods, thereby aligning digitalization and AI-based modeling with sustainable and reliable boiler operations.
Jingjie et al. (Mon,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: