Adaptive PID Control of Furnace Temperature in MSWI Process Based on Reinforcement Learning

This paper tackles the challenge of effectively regulating furnace temperature in the municipal solid waste incineration (MSWI) process. It proposes an approach that integrates reinforcement learning with traditional PID control. Firstly, an analysis of the characteristics of furnace temperature control is conducted to identify the key manipulated variables affecting it. Subsequently, a control strategy specifically tailored for furnace temperature is developed, employing online adaptive PID with a BPNN-fitted actor-critic network (BPNN-ACN-PID). Finally, the effectiveness of the proposed method is validated through control experiments using actual process data from a MSWI plant in Beijing.

I. Introduction

The production of municipal solid waste (MSW) in China has experienced a sharp increase [1], resulting in the phenomenon of "garbage surrounding cities" in numerous urban areas [2] and causing severe pollution to urban soil, water bodies, and air. MSW incineration (MSWI) technology has garnered widespread attention globally due to its benefits in waste reduction, resource utilization, and environmental harm mitigation [3]. Its development in developed countries such as Europe, the United States, and Japan has matured significantly. Due to the uncertainty in composition, large variations in calorific value, and significant fluctuations in moisture content of MSW in China [4], the implementation of automatic combustion control (ACC) systems has been challenging. Currently, MSWI processes in China predominantly rely on manual control by domain experts [5]. Due to the randomness and experience disparities among domain experts, the MSWI process encounters challenges in maintaining stable operations over prolonged periods. Hence, it is imperative to investigate tailored furnace temperature control strategies that account for the specific circumstances in China.