Fuzzy Broad Model Predictive Control Based on Hybrid-driven and Gradient Optimization

Model predictive control (MPC) is an advanced process control strategy widely applied across various industrial processes. Although deep neural networks have been used to enhance traditional MPC performance, they often suffer from high computational complexity and the risk of overfitting. While the application of conventional particle swarm optimization (PSO) in MPC offers global search capabilities, it struggles to meet real-time control requirements due to excessive computational overhead and strong dependency on initial solutions. To address these challenges, this paper proposes a novel fuzzy broad model predictive control approach based on hybrid-driven and gradient optimization. Firstly, an interval type-2 fuzzy broad learning system is employed to construct the predictive model, thereby enhancing nonlinear modeling and uncertainty handling capabilities. Secondly, during the rolling optimization process, a hybrid strategy combining gradient descent and PSO is introduced to ensure fast convergence while improving global search performance. In addition, a knowledge-data-driven surrogate model is built by leveraging the system sample database and particle archive database to significantly reduce computational consumption. Finally, a baseline solving strategy for manipulated variables is designed to improve the safety and reliability of control outputs. The effectiveness of the proposed method is verified through simulation experiments on typical nonlinear systems and actual municipal solid waste incineration process.