With the widespread application of lithium batteries, the focus on battery performance evaluation and safety has increased. Accurate prediction of battery degradation and remaining useful life is now a crucial research topic. Scholars have improved prediction accuracy by integrating multiple parameters related to battery degradation or combining various prediction models. However, environmental noise, sensor limitations, and faults can impact the quantity and quality of available feature sequences, rendering these methods partially ineffective. This paper proposes an optimized Transformer prediction model based on a Temporal Feature Enhancement Module (TFEM). By decomposing long-sequence signals and accurately capturing temporal features across different frequency bands through convolution, the model significantly improves the accuracy of future degradation process assessments and remaining useful life predictions of batteries, even with a single prediction model and a limited number of feature sequences. The proposed TFEM exhibits strong adaptability and can be easily integrated into other temporal prediction models to enhance prediction accuracy. The prediction model demonstrates high generalizability and robustness, achieving favorable prediction results even in scenarios involving short sequence predictions or a limited number of feature sequences.