基于代理模型的锅炉燃烧优化不确定性可视分析OA

Data-driven surrogate models offer new opportunities for boiler-combustion modelling;however,conventional surrogates often overlook intrinsic uncertainty.Combustion processes also exhibit strong multi-parameter coupling,time-lag fluctuations,parameter sensitivity,and varying operating conditions.To address these issues,this paper proposes a surrogate model-based visual-analytics framework of uncertainty in boiler combustion optimization.First,a Bayesian Attention-enhanced Encoder-Decoder LSTM(BAED-LSTM)model is used to predict time-series outputs with quantified uncertainty.Next,four complementary modules are designed to analyze distinct uncertainty sources:(ⅰ)multi-view correlation visualizations for aleatory uncertainty,(ⅱ)temporal pattern stability based visualization for epistemic uncertainty,(ⅲ)symbolized bi-clustering visualization for temporal-pattern sensitivity,and(iv)improved Markov chain-based visualization for operating-condition similarity.The above components are integrated into an interactive system named BUCOVis.Case studies on real power-plant data demonstrate that the system effectively uncovers combustion uncertainties and provides a fairly comprehensive solution for boiler-combustion optimization.