Dioxin Emission Concentration Forecasting Model for MSWI Process with Random Forest-Based Transfer Learning

Abstract

The dioxin (DXN) emission concentration is an important indicator related to the stability and harmless operation of municipal solid waste incinerator (MSWI) process, and also one of the key parameters for MSWI process to realize optimal operation control. In the actual MSWI process, the process variables such as furnace temperature, grate speed and air pressure are hundreds. However, the modeling samples containing DXN emission concentrations are small, which makes it impossible to establish accurate DXN emission concentration forecasting model. In this study, a forecasting model construction method by using random forest (RF)-based transfer learning was proposed. At first, initial weights are assigned to the source domain and target domain samples. Then, an RF-based DXN emission concentration transfer learning model is established, and the prediction errors are used as the indicator to calculate the weight adjustment coefficients. Finally, the sample weights are adjusted through iteration loop in terms of increase the source domain instance weights that related to the target domain. The method proposed in this paper can transfer the source domain sample to enhance the prediction performance of the DXN emission concentration forecasting model. The experimental results based on the actual industrial data show the effectiveness of the proposed method.