A robust fuzzy tree method with outlier detection for combustion models and optimization

Abstract

The fuzzy tree (FT) method is a new efficient modeling algorithm. To improve the performance of detecting noise and outliers for most industrial applications, this paper proposes a robust fuzzy called weighted FT (W-FT). A typical nonlinear example is used in the numerical experiments to validate the proposed W-FT. Then, the W-FT is used for building the combustion models, mainly three soft sensor models are established considering boiler efficiency, NOx and SO2 emissions by using the historical data of a circulating fluidized bed (CFB) boiler. Compared with other methods, the W-FT method exhibits more robustness, higher prediction accuracy and better generalization capability. Moreover, in basis of above soft sensor models, three types of optimization strategies are proposed to optimize the adjustable parameters by using the modified fruit fly optimization algorithm. Simulation results validate the effectiveness of the proposed optimization strategies, and further demonstrate the practicability of soft sensor models by W-FT.