Energy-conservation potential in the air-condition water system for subway stations is huge due to its conservative design method. Also, for operation strategy of such systems, the operation modes are formulated with the fixed schedule. This paper presents a data-based optimization method to obtain optimal parameters of the system for feedforward control. The data mining models are established by using the data from energy consumption platform of the refrigerating system. The study utilized the box-plot method, kNN algorithm and k-means algorithm to process and repair original data. Then Artificial Neural Network (ANN) model is adopted to developed the forecasting model to assess load, performance and energy consumption of the system. The input features of the models are determined by the existed models and clustering analysis. The optimal parameters under the conditions of different load-ratio range and ambient thermal environments are calculated via Genetic Algorithm and trained equipment models. And the optimal parameters are applied to establish operation schedule based on feedforward control and response time. The optimal feedforward control method is verified by a validated TRNSYS model. When the parameters are optimized, the water system energy consumption can be save by 9.5% in a cooling season.

地铁车站空调水系统由于其设计方法保守，节能潜力巨大。此外，对于此类系统的运行策略，运行模式是按照固定的时间表制定的。本文提出了一种基于数据的优化方法，以获得系统的最佳参数进行前馈控制。利用制冷系统能耗平台的数据建立数据挖掘模型。本研究利用箱线图方法、kNN算法和k-means算法对原始数据进行处理和修复。然后采用人工神经网络（ANN）模型开发预测模型来评估系统的负载、性能和能耗。模型的输入特征由已有模型和聚类分析决定。通过遗传算法和经过训练的设备模型计算不同负载比范围和环境热环境条件下的最佳参数。并根据前馈控制和响应时间应用最优参数来建立运行计划。最佳前馈控制方法通过经过验证的 TRNSYS 模型进行验证。优化参数后，在一个冷却季节，水系统能耗可节约9.5%。