Abstract: Accurate and effective estimation method for the state of health(SOH) of lithium batteries is the key concern in the development of battery management system. It is difficult to accurately estimate SOH due to the measured noise. A method to estimate the SOH of lithium batteries based on isobaric energy analysis and convolutional neural network(CNN)-gated recurrent unit(GRU)-multi-headed attention(MHA) is proposed. Firstly, the relationship between battery energy and voltage in the constant current charging stage is analysed to plot the iso-voltage energy curve. Secondly, its peaks are extracted as the health factor to map the SOH degradation. Finally, CNN is used to extract the potential features, the GRU-MHA-based SOH estimation model is constructed. The results show that the proposed method can effectively overcome the measurement noise, which the SOH estimation error is less than 1%. Meanwhile, the outcome of comparison experiment indicates that the proposed method has better estimation effect.