This study addresses the problem of voltage variations in active low-voltage distribution networks caused by distributed photovoltaic (PV) generation. Three strategies based on model predictive control (MPC) are introduced to flatten the voltage profile in a cost-optimal way. The compared strategies are the business as usual approach that manipulates a controllable on-load tap changer at the primary substation, the problematic feeder control strategy (CS) that adds an additional degree of freedom by controlling the critical secondary substations (SSs), and finally the compensation strategy, which controls the primary substation and compensates the non-critical SSs. A sensitivity analysis on the CSs has been conducted comparing the voltage variation reduction and the asset utilization with regard to the accuracy of the prediction models and the forecasted disturbance data. The results show that better (and more costly) characterisation of these parameters only provide a marginal improvement in the reduction of the voltage variations due to the restriction caused by the heavy tap change penalisation. Moreover, the tested case-study shows that the problematic feeder CS outperforms the compensation strategy in terms of larger voltage variation reduction for similar asset utilisation.