Fast flow regime recognition method of gas/water two-phase flow based on extreme learning machine

Gas/water two-phase flow is widely encountered and of great importance in manufacture process and scientific researches, and recognition of its flow regimes is significant to the accurate measurement of its process parameters. Many groups have been working on the online recognition of flow regimes, but the recognition speed becomes a growing concern for online recognition. It is essential to look for ways to improve the recognition speed of the flow regimes. An efficient algorithm named extreme learning machine is applied to identify the flow regimes of gas/water two-phase flow in this paper. The flow parameters are obtained from ring-shaped conductance sensor, and five features that reflect the characteristics of flow regimes are extracted from the measured data. Based on the extracted features, extreme learning machine, least-square support vector machine (LS-SVM) and backpropagation neural network (BPNN) are adopted to separate the flow regimes. The results show that ELM is capable to recognize the flow regimes with high accuracy, and its recognition speed is faster than the other two popular methods.