High Performance Computing (HPC) is conducted in large computer clusters to support complex scientific applications and simulations. These infrastructures are notoriously power-hungry, creating a pressing need to optimise their usage. Additionally, modern clusters are often considered heterogeneous as they consist of groups of nodes with varying characteristics. This article presents a novel scheduling algorithm, the Heterogeneous Energy-Aware Pairing Scheduler (HEAPS), which aims to reduce energy consumption by performing an a priori estimation of the energy consumption of jobs across available nodes to achieve optimal job-node pairings. The evaluation compares its behaviour to that of different schedulers for homogeneous and heterogeneous clusters, demonstrating that certain configurations of HEAPS can reduce energy consumption in various cluster types and sizes, and in several cases, improve makespan and energy efficiency.