Operation Performance Assessment of Industrial Process Based on PSFA-kDPC Fusion WOA-XGBoost Model

Industrial processes often operate under various operating conditions, which can lead to changes in operation performance. This variability may result in numerous misjudgments when using traditional methods that solely evaluate performance as either ‘good’ or ‘bad’. Therefore, this paper proposes an operational performance assessment methodology based on probabilistic slow feature analysis with k-nearest neighbors density peaks clustuering (PSFA-kDPC) and whale optimization algorithmextreme gradient boosting (WOA-XGBoost). Initially, the PSFA-kDPC is proposed to extract the slow features by maximizing the posterior distribution and to cluster them into different operational conditions. Subsequently, the WOA-XGBoost is employed to train the operational performance assessment model, which categorizes performance into three groups, including ‘optimal’, ‘medium’ and ‘poor’. Finally, to mitigate the potential influence of individual samples on the results, a moving window method is adopted in lieu of single samples, and a forgetting factor weighting strategy is incorporated to compute the mean of the samples within the window. The proposed methodology is applied to an actual furnace negative pressure system in a thermal power plant, with verification results demonstrating its effectiveness in differentiating between various operating conditions.