The state of health (SOH) can usually be described by an exponential model. Traditional least squares and gradient descent algorithms are inefficient for such a special model. In this paper, a Volterra series model is used to approximate the exponential SOH model, where some of the collected data are contaminated by outliers. To alleviate the harmful effects caused by the outliers, a weighted least squares algorithm and a weighted gradient descent algorithm are proposed. The algorithms can adaptively ignore those data which are contaminated by outliers, and only utilize the normal data to identify the Volterra model. Compared with the traditional algorithms, the proposed methods have the following advantages: 1) approximated model has a more simple structure; 2) proposed algorithms can adaptively ignore those data which are contaminated by outliers. A simulation example demonstrates the effectiveness of the proposed methods.