Soft sensing modeling of dioxins for waste incineration based on small data sets

Since the online measurement of Dioxins during waste incineration is difficult, it could only be analyzed offline with small samples obtained. Aimed at this problem, a novel soft sensing methodology that can be well generalized is studied. Firstly, bootstrap resampling approach and noise injection are performed for small samples in order to increase the amount of the samples and improve the diversity. Then, the information entropy is introduced to the error rule function for the unknown distributing of original samples and a neural network with the maximum entropy is constructed. Finally, a soft sensing regression model of dioxins is built based on the entropy neural network. Simulation results show that this model has a high precision and a good ability of generalization. The mean and maximum of relative error between actual and predicted values are 0.167% and 1.21%, respectively. This method provides a reference for detecting dioxins online during waste to energy.