Power capping is an important solution to keep the system within a fixed power constraint. However, for the power-constrained systems, powercap needs to be reasonably allocated according to the workloads of compute nodes to achieve trade-offs among performance, energy and powercap. Thus it is necessary to model performance and energy and to predict the optimal powercap allocation strategies. Existing power allocation approaches have insufficient granularity within nodes. Modeling approaches usually model performance and energy separately, ignoring the correlation between objectives, and do not expose the Pareto-optimal powercap configurations. Therefore, this paper proposes an approach to predict the Pareto-optimal powercap configurations on the power-constrained system for MPI and OpenMP parallel applications. It uses the elaboratly designed micro-benchmarks and a small number of existing benchmarks to build the training set, then applies a multi-objective machine learning algorithm which combines the stacked single-target with extreme gradient boosting to build multi-objective models. The models can be used to predict the optimal PKG and DRAM powercap settings, helping compute nodes perform fine-grained powercap allocation. Compared with the reference configuration, our models can achieve an average powercap reduction of up to 48%, average energy reduction of up to around 20%, with only 10%~30% maximum performance drop.