Long-term battery degradation prediction is an important problem in battery energy storage system (BESS) operations, and the remaining useful life (RUL) is a main indicator that reflects the long-term battery degradation. However, predicting the RUL in an industrial BESS is challenging due to the lack of long-term battery usage data in the target’s environment, domain difference between BESS environments, and incomplete battery charging/discharging patterns in industrial scenarios. To address these challenges, we propose a retrieval-based approach, which predicts the RUL of the target battery based on the full-lifetime usage data of reference batteries retrieved from other environments. The basic idea is that the reference batteries with common early-life features are more useful for predicting long-term degradation of the target battery. Based on experiments with both laboratorial datasets and industrial datasets, our method can constantly achieve higher prediction accuracy than state-of-the-art baselines.