A time-varying optimization strategy for battery cluster power allocation is proposed to minimize energy loss in battery energy storage systems (BESS). First, the time-dependent loss characteristics of both storage and non-storage components in BESS are analyzed. Based on this analysis, steady-state and transient methods for evaluating battery loss are proposed. Second, considering the distinct time-varying characteristics of various BESS components, the load-rate vs. equivalent-efficiency curve and the current-loss power component gradient field are introduced as analytical tools. These tools facilitate the derivation of optimization path for both time-varying and time-invariant energy components of BESS. Building on this foundation, a time-varying optimization strategy for battery cluster power allocation is developed, aiming to minimize energy loss while fully accounting for the dynamic characteristics of BESS. Compared to real-time optimization, this strategy prioritizes global optimality in the time domain, mitigates the risk of dimensionality curse, and enhances BESS efficiency. Finally, a Simulink/Simscape model is established based on real-world data to simulate internal component losses within BESS. The effectiveness of the proposed strategy is validated under a peak shaving scenario. Results indicate that, after optimization, the annual operational loss of BESS is reduced by 2.40%, while the energy round-trip efficiency is improved by 0.59%.