A research about the predictive control of dynamic feedforward neural network based on particle swarm optimization

Lizheng Liu^{1, 2}, Fangai Liu², Feng Yang¹ 

¹Shandong University of Finance and Economics, Jinan, Shandong Province, China

²Shandong Normal University, Jinan, Shandong Province, China

The paper proposes the Dynamic Feedforward Neural Network based on Hidden Particle Swarm Optimization (HPSO-DFNN) to deal with the model predicative control problem of unknown nonlinear delay systems. It realizes quick, precise system modelling for controlled objects. Besides, the Smith predictive double controllers are designed to separate fixed set point control from external disturbance. The DFNN based on large-scale PSO is treated as an identifier and a predictor for the complex controlled objects with the purpose of increasing the robustness of the control system. Furthermore, aiming at the problem of constrained multi-input-multi-output (MIMO) model predictive control, rolling optimization is conducted to obtain controlled quantity through the PSO algorithm. After that, a combined neural network structure is put forward and applied to system modelling. Finally, the paper uses the typical nonlinear model to verify its effectiveness.