A hybrid NOx emission prediction model based on CEEMDAN and AM-LSTM

Abstract

Frequent load changes pose a major challenge to the realization of pollutant control in power plants. Accurate and reliable NOx concentration prediction models are of great significance for reducing emissions. In this study, a novel NOx concentration prediction model is proposed. First, the model decomposes original historical data into a set of constitutive sequences using complete ensemble empirical mode decomposition adaptive noise (CEEMDAN) to extract data features. Second, a deep learning network based on an attention mechanism (AM) and long short-term memory (LSTM) is applied to predict each component separately. Finally, the prediction results for each component are integrated to obtain a NOx concentration prediction. Three different datasets from the distributed control system of a coal-fired power plant are acquired to train and verify the proposed hybrid CEEMDAN-AM-LSTM model. Experimental results verify that the CEEMDAN-AM-LSTM model can accurately predict NOx concentration and is superior to many alternative methods.