A learning control solution to the problem of finding a finite-time optimal control history that minimizes a quadratic cost is presented. Learning is achieved without requiring a detailed knowledge of the system, which may be affected by unknown but repetitive disturbances. The objective of this work is to focus on the temperature control of a semi-batch chemical reactor used for fine chemicals production. With less modelling investigation, a feedback-feedforward control structure is proposed for ensuring the tracking performance of the desired temperature profile. Such strategy is derived from the family of the Iterative Learning Control named Batch Model Predictive Control (BMPC). The synthesis of the considered strategy is studied and improvements of the algorithm features are proposed. Experimental results are presented to illustrate the practical appeal and effectiveness of the proposed scheme.