A CNN Approach for Optimal Power Flow Problem for Distribution Network

The OPF problem for distribution networks is always a challenge in the operation or planning of power systems. The computational efficiency is crucial when solving the OPF problem. This paper develops a CNN (Convolutional Neural Network) approach, named ConvOPF, to obtain optimal power flow for the distribution network. Firstly, a CNN is constructed and trained to learn the relationship between the load and the generations. Then, the voltage magnitude and phase angle are computed directly by using an AC power flow solver. This procedure significantly guarantees the power-flow balances and reduces the number of variables that the CNN model needs to predict. Therefore, the proposed ConvOPF can reduce unnecessary iterations and improve computational efficiency significantly. The experimental results of the 33-bus power grid verify the superiority of the proposed ConvOPF approach, which computing time can speed up significantly compared to the conventional iteration-based solvers.