The present disclosure is related to hybrid vehicles. The teachings thereof may be embodied in vehicles as well as operation schemes meant to increase energy efficiency, such as a method comprising: detecting multiple consumption parameters of the hybrid vehicle; determining a future state of charge of a traction battery of the vehicle by mapping the consumption parameters onto a state-of-charge value, wherein the mapping includes classifying the multiple consumption parameters according to trainable class boundaries; training the class boundaries based at least in part on the detected consumption parameters and an associated measured state of charge; and adjusting an operating parameter of a traction power component of the hybrid vehicle according to the determined future state of charge.