Naturally, the resources within the car's battery processor are limited. Thus, it is necessary to find a cheap method in terms of calculation effort. Here, subspaces seem most promising as all they require is matrix multiplication and inversion, all of which can be done cheaply. Neural networks convince by their tight fit, however to implement them efficiently specially designed microchips would be needed. Doing a direct non-linear estimation would not need many changes in the existing method, and starting from the right initial values can produce a significantly lower error, however, to find these initial values so far no efficient procedure has been found.