NOx Emissions Prediction With a Brain-Inspired Modular Neural Network in Municipal Solid Waste Incineration Processes
Xi Meng, Jian Tang, Junfei Qiao
Abstract
The timely and accurate measurement of nitrogen oxides (NOx) emissions is important for efficient pollution controlling of municipal solid waste incineration plants. With the aim to design an efficient and effective prediction model for NOx concentrations, a brain-inspired modular neural network (BIMNN) is developed in this article. First, a biologically inspired modularization technique is proposed in which the topological modularity gives rise to functional modularity. Consequently, different modules correspond to different tasks, improving the network efficiency by performing task decomposition. Subsequently, an adaptive task-oriented radial basis function (ATO-RBF) neural network is applied to construct each module based on assigned subtasks. The ATO-RBF neural network is comprised of a structure self-organizing mechanism and an adaptive second-order learning algorithm, providing basis for learning performance and generalization ability of BIMNN. Finally, during the testing or application stages, a competitive strategy is utilized to select the modules which can be adapted to the current task, aiming to enhance the efficiency of BIMNN. The proposed prediction methodology is verified using industrial data, and the experimental results demonstrate the advantages of the BIMNN-based prediction model on speed and accuracy.