The state of health （SOH） of a lithium-ion battery reflects the aging degree of Lithium-ion the battery. When the battery is charged in constant current-constant voltage mode， the charging curves with different aging degrees are also different. Based on this fact， this paper proposes a data-driven model based on a temporal convolutional network （TCN） to establish the mapping relationship between the charging curve and SOH. TCN is a novel neural network composed of multi-layer causal convolution， which can encode the sequence of sampling points on the charging curve. The experiment proves that the encoding vector is easier to establish the mapping relationship with SOH. The experimental results show that the proposed SOH estimation model has high estimation accuracy andgood adaptability to different types of batteries.