Lithium-ion batteries have been widely used in various aspects of our lives, playing a crucial role in numerous applications. The state of health (SOH) serves as a pivotal indicator, and accurate prediction of SOH is essential for the safe utilization, management, and maintenance of lithium-ion batteries. In order to accurately predict SOH, a hybrid prediction model by combining bidirectional gated recurrent unit (BiGRU) and Transformer with multi-head attention mechanism (AM) is proposed, which can effectively address the challenge of long time series prediction. In the proposed prediction model, the indirect health indicator (HI), which can characterize the degradation of lithium-ion batteries, is fed into the BiGRU to learn the hidden states of the input features and thus further extract time series features. On this basis, multiple attention is given to the Transformer encoder layer and the input feature vectors, which gives it a better performance in the long-term dependence of the time series. The study based on the lithium-ion battery data from NASA Prediction Center of Excellence (PCoE) shows that the proposed BiGRU-Transformer model has higher accuracy, better robustness and generalisation capability.