Computationally Efficient Data-Driven Surge Map Modeling for Centrifugal Air Compressors

For the compressor operation, surge is a detrimental phenomenon of large oscillating pressure and flow in air compressors, which occurs usually at too low flow rate for a given discharge pressure. For many industrial air compressors, surge map is widely used for surge avoidance control. Field operation of centrifugal air compressors for manufacturing plant has shown that the surge map of a compressor may vary dramatically with ambient and operational conditions. In our previous work, data-driven surge map modeling method has been developed to obtain surge maps under different ambient air conditions, where the asymmetric support vector machines (ASVM) were developed for obtaining the surge map models based on actual surge test data. Support vector machine algorithms are generally computationally intensive, which may increase the complexity and cost of implementation. In this paper, a method of effectively selecting the support vectors is applied to the ASVM based surge map modeling framework. The modeling results correctly predict all gathered surge conditions with much less support vectors and lower orders kernel. About four times reduction of model complexity was resulted.