Model NOx Emissions Emitted from Coal-fired Boilers Based on Differential Evolution Fast Learning Network

Abstract

Low NOx combustion optimization is a simple, efficient and inexpensive NOx emission reduction technology for coal-fired power plants. Establishing NOx prediction model is an important part of this technology. Fast learning network (FLN) is an improved neural network proposed in recent years, which is simple and efficient. However, randomly generated hidden layer thresholds and input weights would affect the performance of FLN. To solve this problem, differential evolution (DE) algorithm is employed to optimize hidden layer thresholds and input weights. The proposed model was applied to some 330MW coal-fired boiler together with FLN. Each model was repeated 51 times to consider the randomness of both models caused by the randomness of DE algorithm and the stochastic initialization of the original FLN. Results showed that the model has better generalization ability, retaining the good approximation ability and stability. Besides, optimization process of the proposed model is very fast and fit for online combustion optimization.