Research on Neural Network-Driven Particle Swarm Optimization in a Linear Double Inverted Pendulum System

This paper proposes a neural network-driven particle swarm optimization (PSO) algorithm for optimizing the LQR controller parameters of a linear double inverted pendulum system. The algorithm combines the learning capability of neural networks with the global search ability of particle swarm optimization. A three-layer feedforward neural network guides the particle search while employing adaptive inertia weights and learning factor strategies. Experimental results indicate that compared to traditional PSO, GA, and DE algorithms, the new algorithm significantly improves convergence speed, control accuracy, and energy efficiency. In particular, the new algorithm achieves the same performance as other algorithms with 200 iterations in just 70 iterations. Furthermore, the algorithm demonstrates strong robustness against parameter uncertainties and external disturbances, providing an efficient solution for control optimization in complex nonlinear systems.