In this paper a new approach of bilinear predictive control is present. The approach is based in the Bilinear Generalized Predictive Control (BGPC), strategy that makes use of the time-step quasi-linearised NARIMAX model. In that approach, due to the used model, a prediction error exist, which increases with the prediction horizon, degrading the performance of that controller. Thus, in the present approach, a compensated model is used, whose compensation term depends of the prediction horizon. The algorithm and the results obtained in a example simulation are shown, evidencing that a new approach presents a better performance than the controller based on the quasi-linear model.