A combined method for voltage over-limit prediction with high accuracy in distributed PV-connected systems

As high-penetration renewable energy integration technologies progress, distributed photovoltaic (PV) systems are widely deployed in distribution networks. It is worth noting that the intermittent power output causes the problem of voltage over-limit at the point of common coupling which poses a threat to grid stability and PV efficiency. To address this issue, a novel voltage prediction method integrating complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN), the Newton–Raphson-based optimizer (NRBO), and a bidirectional temporal convolutional network (BiTCN) is proposed. The voltage sequence collected from a typical distributed PV grid-connected system is decomposed into stable sub-signals with CEEMDAN to enhance the interpretability of the original data. The comprehensive feature in the decomposed signal is extracted by BiTCN. To avoid local optimum and ensure the stability of the model, NRBO is introduced to determine hyper-parameters of BiTCN. The results show that the voltage predicted by the proposed method is almost identical to the actual data, and the voltage prediction accuracy of the proposed method is higher than the comparison models. The calculated values of mean absolute error, mean squared error, and decision coefficient (R2) of the proposed method are 0.1035, 0.2472, and 0.9966, respectively. The proposed method is a good candidate for voltage over-limit prediction in the distributed PV grid-connected system. It will help to ensure grid stability and improve power generation efficiency of the distributed photovoltaic systems.