Fusion Prediction Model of Atmospheric Pollutant Based on Self-Organized Feature

Abstract

In response to the new situation and new needs of international industrial modernization transformation, machine learning provides important data basis for monitoring air pollution in industrial parks and can control exhaust emissions in industrial production on a human basis. In all kinds of systems, the time series obtained in practice are generally non-stationary, noisy and complicated data influenced by multiple factors. In this paper, the EMD-FUSION model based on the self-organizing model is proposed. First, the complex original data were decomposed into empirical mode decomposition (EMD). Based on the volatility and periodicity of each intrinsic mode function (IMF), the three subs modal time series with similar change rates were obtained; Secondly, use a back propagation (BP) to predict simpler sub-modal time series data, while the more complicated sub-mode timing series data of the two groups were predicted by the gated recurrent unit (GRU) network. Finally, the prediction results of three sub-series are added and fused to obtain the prediction data of the original data. Through experiments on the atmospheric data of the industrial park provided by Taiyuan environmental monitoring center station, the correctness of the model is verified.

Keywords

• Air pollution in industrial parks
• EMD-FUSION
• machine learning
• self-organized model
• time series