Neuro-fuzzy control of a steam boiler-turbine unit

Abstract

Conceptually, fuzzy logic possesses the quality of simplicity. However, its early applications relied on trial and error in selecting either the fuzzy membership functions or the fuzzy rules. This made it depend rather too heavily on expert knowledge which may not always be available. Hence, a self-tuning or an adaptive fuzzy logic controller (FLC) such as Adaptive Neuro-Fuzzy Inference System (ANFIS) removes this stringent requirement. This paper demonstrates the application of ANFIS to a 160 Mw nonlinear multi-input multi-output (MIMO) steam boiler-turbine unit. The space of operating conditions of the plant is partitioned into five regions. For each of the regions an optimal controller is designed. The resulting five linear controllers are used to train ANFIS. Simulation results showed that the fuzzy controller closely reproduced the optimal performance in each of the design points and surpassed any single linear controller in these operating regions. These results also reveal the robustness of the FLC to parameter variations.