Flue Gas Oxygen Content Modeling Approach Based on BPNN and LRDT in Municipal Solid Waste Incineration Process

Flue gas oxygen content is one of the key controlled variables of municipal solid waste incineration (MSWI) process. And it is closely related to the combustion efficiency and pollutant emission. At present, there are few reports on flue gas oxygen content model as the controlled object. The article proposes a flue gas oxygen content modeling method based on the fusion of back propagation neural network (BPNN) and linear regression decision tree (LRDT). Firstly, the BPNN model is used to construct the main model. Then, LRDT is used to construct the compensation model by using the error of the of main model as the output. Finally, the fusion output is obtained by summarizing the LRDT and BPNN model. Experimental results based on actual operating data of a MSWI plant in Beijing shows the feasibility and effectiveness of the proposed method.

I. Introduction

With the advancement of urbanization, the total amount of municipal solid waste (MSW) in China has grown rapidly[I], resulting in the phenomenon of “garbage siege” in many places in China [2]–[3]. MSW incineration (MSWI) is a widely used MSW treatment method with the advantages of harmlessness, resource utilization and reduction [4]. In China, the MSWI plants mostly rely on domain experts to estimate the combustion state and control pollutant concentration through “air and material distribution” operation [1]. Thus, this manual control mode has the disadvantages such as subjective and unstable [5]. It can lead to unstable operation of the MSWI process and even excessive pollution emissions [6]. Therefore, it is particularly important to make intelligent optimal research for key controlled variables such as furnace temperature and flue gas oxygen content etc. Flue gas oxygen content is closely related to the combustion efficiency and pollutant emission concentration in MSWI process. The focus of this article is to construct a data-driven controlled object model.