Event-Triggered Adaptive Model Predictive Control of Oxygen Content for Municipal Solid Waste Incineration Process

Oxygen content in flue gas is a key variable in the operation of the municipal solid waste incineration (MSWI) process. However, the control performance of oxygen content could not be guaranteed due to the inherent nonlinearity and uncertainty of the MSWI process. Furthermore, frequent operations of actuators in the conventional time-triggered system may also increase the computational burden and energy consumption. In this study, an event-triggered adaptive model predictive control (ET-AMPC) scheme is developed to solve the above problems. First, an improved long short-term memory (ILSTM) neural network is designed to construct the prediction model, in which the parameters are determined by the particle swarm optimization (PSO) algorithm. Besides, during the control process, the model parameters can be adjusted adaptively by an online updating strategy to tackle the uncertainty. Second, an event-triggered strategy is proposed to reduce the computational burden and energy consumption, wherein the control laws are updated only when certain triggering conditions are satisfied. Third, the gradient descent method is applied to obtain the optimized control law. Moreover, the convergence of the prediction model and the stability of the whole ET-AMPC are analyzed. Finally, the proposed ET-AMPC scheme is evaluated by real industrial data. The experimental results demonstrate that the ET-AMPC scheme can achieve satisfactory tracking control performance with fewer triggering events. Note to Practitioners—The proposed ET-AMPC is an intelligent control scheme of oxygen content for the MSWI process. The proposed control scheme has adequate tracking performance with varied set-points and less computational burden. In practice, practitioners should obtain accurate historical input and output data, and preset event-triggered thresholds. The hyperparameters of the LSTM neural network are automatically determined by the PSO algorithm, and the uncertainty can be captured with the online updating strategy. The accurate prediction model is beneficial for improving the control quality. Moreover, the optimized control law is obtained with the intuitive gradient descent method. In addition, the event-triggered strategy can further reduce unnecessary control optimization operations with better practicality and ease of usage. The superiority and practicability of the proposed method are verified against actual industrial data.