An Improved GA-KRR Nested Learning Approach for Refrigeration Compressor Performance Forecasting

Abstract

The long duration of refrigeration compressor performance tests is a key factor restricting the quality testing efficiency and the delivery times. To reduce the time of quality tests in the refrigeration compressor manufacturing systems, data-driven technology is used for forecasting the compressor performance using unsteady-state data in early test phase. The typical methods usually encapsulate two distinct blocks: input range selection and performance prediction. Such fixed and hand-crafted input range, which is crucial for the prediction accuracy and test time saving, may be a sub-optimal choice for diverse varieties of the compressors and prevent their usage for real-time applications. In this paper, we proposed a compressor performance forecasting approach using GA-KRR (genetic algorithm - kernel ridge regression algorithm) nested learning that has a heuristic design to automatically hunt the best input range and a nested learning design to fuse the automatic input range selection and performance prediction into a single learning body. The experimental results on real-world data show the outstanding performance of proposed approach compared with relative approaches, which indicates the test time can be reduced 75%.