In systems such as photovoltaic-energy storage power generation and DC microgrid, isolated triple active bridge(TAB) with higher power density and lower cost are often used. Power coupling issue is critical for TAB, which could be classified into look-up table based and artificial neural network (ANN) based approach. However, these methods face the problem of large memory consumption or poor decoupling effect under wide operation range. Therefore, this paper proposes an optimized ANN based decoupling method by data linear fitting and selecting. A set of optimized process is given and its effectiveness is verified by Matlab/Simulink simulation. The results show that the proposed method achieves a good trade-off between data storage memory and decoupling effect under varying working conditions.