Synthesis of a Nonlinear PID-Controller based on a Neural Network Representation for Adaptive Control of Dynamic Objects
Ochiq kirish: yoʻq
Annotatsiya
This study introduces an innovative approach to constructing neuro-PID controllers utilizing neural network architectures for regulating dynamic systems with time-varying characteristics. The proposed framework reconceptualizes the discrete PID structure as a two-layered artificial neural network, wherein the initial layer incorporates nonlinear activation mechanisms while the subsequent layer maintains linear transfer properties. The synthesis methodology unfolds through distinct stages: preliminary computation of standard PID parameters to establish second-layer connection weights, followed by optimization of first-layer nonlinear activation profiles to elevate transient performance metrics. Global search across parameter domains employs genetic evolution strategies, ensuring solution discovery within multimodal optimization landscapes. Simulation analyses reveal that replacing traditional linear PID architectures with the developed neuro-PID framework yields significant performance enhancements for nonlinear process control, manifested through minimized steady-state errors, reduced overshoot percentages, and shortened settling durations. The methodology preserves the intuitive structure and design transparency inherent to classical PID control, positioning it as a practical solution for modernizing industrial control infrastructure across manufacturing operations, power generation plants, and related automation sectors.
Hali tarjima qilinmagan