Graph network-based design optimization of multiple chiller systems with knowledge from existing systems

Operation data from existing multiple chiller systems contain valuable knowledge, which can guide chiller configuration for new systems. However, conventional strategies are often limited to comparing a few proposed configurations, failing to exploit the valuable knowledge. To address this, the graph network-based design optimization strategy is developed. The graph network consists of chiller nodes, part load ratio (PLR) nodes, synergy edges and efficiency edges. Specifically, synergy edges and efficiency edges are the key knowledge that whether chillers can operate efficiently when they are operated simultaneously and whether a chiller can operate efficiently under certain PLR, respectively. Based on the graph network, the optimal configuration is developed. First, synergy edges screen for feasible configurations. Second, efficiency edges identify candidate configurations that maintain high efficiency across all typical loads. Finally, estimated annual energy consumption determines the optimal configuration. To assess its effectiveness, the strategy is applied to a case study in a lithium battery factory. Validation results indicate that the optimal configuration obtained from the proposed strategy improves energy efficiency by 5.94% relative to the original configuration. Compared with other configurations, each chiller in the optimal strategy operates efficiently under more typical loads. To further validate the strategy, configurations selected by conventional strategies are used for comparison, showing that the proposed strategy achieves a 4.49% improvement over the best configuration obtained by conventional strategies. This study establishes a graph network-based optimal configuration that enables the intelligent transfer of experience from existing systems to new designs.