With advancements in electric power systems, AC optimal power flow (AC-OPF) has become crucial for ensuring operational reliability and enhancing the efficiency of power system management. Since AC-OPF models are characterized by non-convexity, most existing methods struggle to guarantee convergence to the global optimal solution. This paper presents a formulation of a global optimization problem for AC-OPF, incorporating real-world physical components such as volt-ampere reactive compensators and transformers. An approach based on collaborative neurodynamic optimization (CNO) is developed to solve the AC-OPF problem. The scattered searches of multiple projection neural networks with particle swarm optimization update rule are employed to seek the global optimum. Experimental results demonstrate that the CNO-based approach is both effective and superior to eight existing AC-OPF methods.