Heterogeneous parallel machines with peak energy consumption constraints are associated with stochastic customer order scheduling problems, i.e., the total energy consumption of all machines must not exceed a given threshold. Thus, in this study, we propose an algorithm based on model predictive control to minimize the expected cycle time of customer orders. The proposed algorithm effectively allocates energy, thus improving equipment efficiency within peak energy constraints. It also manages proper rolling scheduling of demands to deal with uncertainties in incoming customer orders. The model is then theoretically analyzed with respect to optimization strategy, order cycle time, product difference, and machine speed. Our experiment results verify the effectiveness of the proposed algorithm and excavate useful managerial insights to guide practical applications better.