In this paper, we consider controller synthesis problem of power assisting system for almost-periodic motions. The simplest assist method would be to apply assist force generated by actuators in proportion to the instantaneous value of the force generated by human. There seems to be no other choice when the motion is irregular and unpredictable. However, human persistent tasks are sometime periodic and it can be shown that such a strategy is not optimal in the sense of energy efficiency. In our previous works, an optimality condition for periodic motion is derived. This result implies that to flatten the force pattern by removing the input pulsation improve the efficiency. Thus, optimal power assisting control methods for periodic motions based on a frequency shaping were investigated. However, we observed that the pedaling frequency of human is not constant and fluctuates to some extent. In order to deal with such a situation, an adaptive compensation is introduced here. In particular, we propose a power assisting control with an adaptive notch filter to suppress the velocity pulsation. The effectiveness of the proposed assisting control method is verified through numerical simulations and experiments by using a commercial electric power assisted bicycle. Finally, the energy efficiency is evaluated by monitoring voltage drops of the battery during our experiments.