Adaptive wavelet neural network for prediction of hourly NOX and NO2 concentrations

Adaptive neural network is a powerful tool for prediction of air pollution abatement scenarios. But it is often difficult to avoid overfit during the training of adaptive neural network. In this paper, based on the wavelet theory, an algorithm is proposed to improve the generalization of adaptive neural network during online learning. The algorithm trains adaptive wavelet neural network to model hourly NO_X and NO₂ concentrations of variance of emission sources. Results show that the algorithm improves the generalization and the convergence velocity of adaptive wavelet neural network during online learning. The simulations also illustrate that adaptive wavelet neural network is capable of resolving variance of emission sources.