Due to the inevitable degradation of Lithium-ion batteries (LIBs) during its lifetime, remaining useful life (RUL) prediction methods are adopted for ensuring the stable and safety operation of electrical equipment. To make up the deficiencies of single model-based or data-driven prediction approach, this paper proposes a new hybrid framework using a weight optimization unscented Kalman filter (WOUKF) and attention based bi-directional long short-term memory (BiLSTM-AM). To be specific, firstly, a Fourier model is proposed to describe the degradation of LIBs instead of the traditional double exponential model. Then, a WOUKF algorithm is designed to identify the model parameters efficiently even in the case of poor initialization. Next, the trends of prediction residual between the WOUKF and the true capacity is established by BiLSTM-AM, which is fed back to the WOUKF for updating model parameters during the prediction period. In addition, an error compensation scheme is developed to further improve prediction performance. Finally, the effectiveness of the proposed prognosis framework is verified on two battery datasets. The simulation results show that the proposed method has better prediction accuracy. The maximum root mean squared error (RMSE) and mean absolute error (MAE) of the proposed hybrid framework are 3.5% and 3%, respectively.