Abstract: In the current analysis process of NOx emissions from waste incineration plant, convolutional neural networks are mainly used to predict NOx emissions. Due to the fact that variable data features can only be captured from the spatial dimension, the prediction results have relatively large errors. Therefore, a prediction model for NOx emissions from waste incineration plant based on spatiotemporal convolutional network optimization was proposed. By mutual information analysis, the key process variables strongly correlated with NOx emissions were screened out. The persistent time series data were transformed into multiple intrinsic mode function components through ensemble empirical mode decomposition,extracting multi-scale features and suppressing noise. An intelligent prediction model was established based on multi-scale convolutional neural network and long short-term memory network. By capturing the temporal and spatial variation characteristics of the data, the high-precision prediction of future NOx emission concentrations was achieved. The experimental results showed that the relative error of the model prediction could remain stable within 3%, demonstrating its superior performance. This model not only enhances the accuracy of emission prediction, but also provides reliable data support for the real-time optimization and emission regulation of the incineration process.

Key words: spatiotemporal convolutional network optimization, waste incineration plant, NOx emissions, integrated empirical mode decomposition, mutual information, predictive model