Battery energy storage system expands the flexibility of the electricity grid, which facilitates the extensive usage of renewable energies in industrial applications. In order to ensure the techno-economical reliability of the battery energy storage system, managing the lifespan of each battery is critical. In this paper, a novel evolutionary framework is proposed to estimate the Lithium-ion battery state of health, which uniformly optimizes the two key processes of establishing a data driven estimator. The features in the degradation process of a battery are conveniently measured by a group of current pulses, which last only few seconds. The proposed evolutionary framework selects the most efficient combination of the short-term features from the current pulse test, and guarantees an optimal training process simultaneously. A hybrid encoding technology is applied to mix the feature extraction and the parameters of support vector regression in one chromosome. With the benefit of the proposed evolutionary framework, the battery state of health is estimated by using support vector regression and genetic algorithm in a more efficient way. A mission profile corresponding to batteries providing the primary frequency regulation service to the power system is used to cycle two Lithium-ion batteries for the validation of the proposed method.