The objective of this study is to improve the performance of the extremum-seeking control ( ${\rm{ESC}}$) technique in terms of time and accuracy of convergence toward the optimum operating point of a dynamic system subject to the effect of external disturbances. More precisely, the idea is to reduce the undesirable effect of time scale separation in ${\rm{ESC}}$ on the performance of the closed loop system. The method consists in adaptively controlling the excitation signal amplitude using a neural network (NN) model, which gives a real-time estimate of the optimal operating point based on the measurement of the external disturbances. Stability of the proposed ${\rm{ESC}}$ with adaptive excitation, referred to in the following as ${\rm{ES}}{{\rm{C}}_a}$, is demonstrated. The superiority of ${\rm{ES}}{{\rm{C}}_a}$ compared to ${\rm{ESC}}$ in terms of accuracy and time of convergence to the optimum is demonstrated both theoretically and experimentally, in the case of the optimization of a photovoltaic panel system.