To address the issues of high energy consumption and inadequate optimisation during the design phase of large-scale building chillers, this paper proposes a multi-constraint combinatorial optimisation method for chillers.First, exhaustive generation produces feasible combinations satisfying design load constraints; based on the COP-PLR curve, a discrete energy consumption model for individual units is constructed. Using Min-plus convolution, the minimum power-load lookup function for each combination is obtained through unit-by-unit superposition. These are mapped onto hourly load sequences from a typical meteorological year to derive hourly electricity consumption. A dynamic programming recursive model evaluates energy consumption versus switching frequency during the cooling season under constraints including start/stop cycles, switching frequency, and energy-saving thresholds. An energy efficiency-stability Pareto frontier is constructed to select the optimal combination.Case studies demonstrate that this method can efficiently identify optimal combinations that balance energy efficiency and stability, with constraint parameters capable of altering the selection of optimal combinations.