Transformer is the core equipment of power system and the key to safe and reliable operation. It is very important to diagnose transformer discharge fault timely and effectively. Due to its remarkable advantages, DGA (Dissolved gas analysis) plays a decisive role in fault diagnosis of transformer. However, the severity of discharge fault can't be assessed effectively by the traditional DGA method. Therefore, an improved method named ETDCG (energy-weighted total dissolved combustible gases) is proposed based on the thermodynamics theory in this paper. Firstly, the energy-weighted factors were introduced to characterize the energy difference of typical gases and the ETDCG can be calculated based on this enthalpy changes. Then, an experimental platform for transformer discharges was built. Finally, the experiments of different discharge models and different discharge stages were carried out and the correlation between the discharge energy and ETDCG is verified.