Extremum seeking control (ESC) is a model-free control solution for real-time optimization of system operation where model acquisition is difficult and/or cost prohibitive. For many HVAC and refrigeration systems, there can be a large number of candidate inputs for ESC design; however, some inputs affect the performance measure to a greater degree than others. This article presents an online input selection method for multivariable ESC, which uses a singular value decomposition (SVD) analysis coupled with a dither-demodulation-based online Hessian estimate for the underlying static map. A subset of physical inputs or a new set of inputs via linear combination of the physical inputs can be determined using the proposed approach. We present an analysis for quantifying the loss bound of achievable optimum output with the underlying input selection. Further, the Hessian estimation error bound is quantified with perturbation analysis. The proposed method is evaluated with Modelica simulation models of chilled-water plants, one with a single chiller and the other with two parallel chillers. The simulation results validate the effectiveness of the proposed method of input selection.

极值搜索控制 (ESC) 是一种无模型控制解决方案，用于实时优化模型获取困难和/或成本过高的系统操作。对于许多 HVAC 和制冷系统，可能有大量的候选输入用于 ESC 设计；但是，某些输入比其他输入对绩效衡量的影响更大。本文介绍了一种多变量 ESC 的在线输入选择方法，该方法使用奇异值分解 (SVD) 分析以及基于抖动解调的在线 Hessian 估计来对基础静态地图进行估计。可以使用所提出的方法确定物理输入的子集或通过物理输入的线性组合的新输入集。我们提出了一种分析，用于量化可实现的最佳输出与基础输入选择的损失界限。此外，Hessian 估计误差界限通过扰动分析进行量化。所提出的方法通过冷冻水设备的 Modelica 模拟模型进行评估，一个带有单个冷却器，另一个带有两个并行冷却器。仿真结果验证了所提出的输入选择方法的有效性。