Abstract: The concept of subnetwork and the strategy to magnify experiment data with an appropriate multiple were proposed based on the characteristic of the phase equilibrium data of the aqueous two-phase system formed by ［C4mim］Br/K3C6H5O7. Then the neural network model was established by dividing the collected data at  278.15 K, 298.15 K and 318.15 K into training and validation samples. The suggested model was slightly dependent on experiment and outperformed the Setschenow-type equation in terms of calculation accuracy