Abstract: In view of the problem of random and continuous data missing in municipal solid waste incineration (MSWI) process, a filling method of missing data based on expert experiences and ensemble model of reduced features was proposed. First, according to the lack conditions of process data, the missing data were divided into three types, i.e., missing with random distribution, time dimension and feature dimension. Then, the former two ones were filled based on expert experience, and the distribution similarity and mutual information (MI) correlation were used to select the modeling data and reduce the input feature for the third type one. Sub-models with complementary characteristics based on random forest (RF), gradient boosting decision tree (GBDT) and back propagation neural network (BPNN) were used to predict the preliminary missing values. Furthermore, the fusion model based on Bayesian linear regression (BLR) was used to obtain the final data filling values. Finally, a soft-sensor model of dioxin (DXN) emission concentration based on deep forest regression with cross-layer full connection (DFR-clfc) was established to verify the filling effect. Results show that the proposed method improves the data quality of MSWI process.