<p>Prediction of the lithium-ion battery (LIB) degradation trajectory is crucial for new energy vehicles. In recent years, bi-directional long short-term memory (Bi-LSTM) networks have achieved good performance in LIB life prediction because they can effectively capture the long-term dependencies of time series data. However, Bi-LSTM has a large number of parameters, high complexity, and low operational efficiency, making it difficult to deploy on an edge device with limited memory and computational power, which greatly restricts its application scenarios in the era of industrial Internet of things. To this end, we propose a knowledge distillation-based network compression framework for degradation trajectory prediction of LIB. In our framework, a high-complexity network is first leveraged to sufficiently extract LIB nonlinear degradation knowledge. Then, a knowledge distillation (KD) technique is integrated to guide the learning of a lightweight heterogeneous network, thereby realizing knowledge transfer between different network structures. The effectiveness of our proposed framework is validated on the Maryland battery dataset. Extensive experimental results indicate that our proposed method outperforms other benchmarks in terms of model complexity and prediction accuracy. Specifically, when the parameter volume of the KD-based lightweight network shrinks by a factor of 155.3, it even achieves comparable performance to the intricate teacher network.</p>