Local Characteristics Preservation Based State Identification in Oil-gas-water Three-phase Flow Process

Oil-gas-water three-phase flow process is a dynamic non-stationary system, and the flow states are extremely complicated due to the three-phase mixture. Aiming at state identification of water-based dispersed three-phase flow, a strategy combining locality preserving projections (LPP) and Gaussian-weighted k-nearest neighbor (GWKNN) is proposed. Based on statistics pattern analysis (SPA), rich state information can be obtained. By constructing the combined model of LPP and GWKNN, the local neighborhood structures are preserved and the local characteristics of flow states are highlighted. The effectiveness of the strategy is demonstrated by experiments.