An Adaptive Extreme Learning Machine for Modeling NOx Emission of a 300 MW Circulating Fluidized Bed Boiler

Abstract

Extreme learning machine (ELM) provides high learning speed, but generalization performance needs to be further improved. Therefore, we propose an adaptive ELM with a relaxation factor $\lambda$ (A-ELM). In A-ELM, according to the nonlinear degree of actual data, the output layer obtains adaptively $1-\lambda$ rate information through the hidden layer and $\lambda$ rate information through the input layer. Since the relaxation factor $\lambda$ is bound up with the input weights and hidden biases of A-ELM, in order to obtain the optimal $\lambda$, $\lambda$, input weights and hidden biases are obtained together by teaching–learning-based optimization (A-ELM-TLBO). Then, 15 benchmark regression data sets verify the performance of A-ELM-TLBO. Finally, A-ELM-TLBO is applied to set up the mapping relation between NOx emission and operational conditions of a 300 MW circulating fluidized bed (CFB) boiler. Compared with six other models, experimental results show that A-ELM-TLBO has good approximation ability and generalization performance. So, A-ELM-TLBO provides a good basis for tuning CFB boiler operating parameters to reduce NOx emission.