Abstract
Learning an optimal control strategy from the optimized operating dataset is a feasible way to improve the operational efficiency of HVAC systems. The operation dataset is the key to ensuring the global optimality and universality of the operation strategy. Currently, the model-based method is commonly used to generate datasets that cover all operating scenarios throughout the cooling season. However, thousands of iterative optimizations of the model also lead to high computational costs. Therefore, this paper proposed a scenario reduction method in which similar operating scenarios were grouped into clusters to significantly reduce the number of optimization calculations. First, k-means clustering (with dry-bulb temperature, wet-bulb temperature, and cooling load as features) was used to select typical scenarios from operating scenarios for the entire cooling season. Second, the model-based optimization was performed with the typical scenarios to generate the optimal operating dataset. Taking a railway station in Beijing as a case study, the results show that the optimization time for the typical scenarios was only 1.4 days, which was reduced by 93.1% compared with the 20.6 days required to optimize the complete cooling season scenario. The optimal control rules were extracted, respectively, from the above datasets generated under the two schemes, and the results show that the deviation of energy saving rate was only 0.45%. This study shows that the scenario reduction method can significantly speed up the generation of the optimal control strategy dataset while ensuring the energy-saving effect.