The transient fault-current is studied using wavelet analysis combined with entropy. The energy entropy is defined as universal applicable parameter which measures the degree of uncertainty and complexity of fault transient quantitatively. The energy entropy measure and saturation entropy are defined to make statistic of the fault energy rules according to analyzing the characteristics of spectrum energy distributions. The multi-scale energy entropy measure is found and used to cluster fault models. Then， the single-phase-to-ground fault is divided into three classes: strong fault， small angle fault and weak fault， based on which the adaptive fault line selection method is proposed. The effectiveness of proposed method is proven by simulation and fault recording data.