Data Driven Approach for Optimal Power Flow in Distribution Network

Optimal power flow (OPF) provides solutions to power flow problem like network loss minimization and voltage profile improvement in the distribution network. This task is achieved by the optimal setting of control variables subjected to operational constraint. Traditionally, OPF solutions are calculated by model-driven approaches which are based on mathematical assumption. The accuracy of these model based depends on the theoretical concept like convexity, differentiability and continuity. Accuracy of the model is subjected to system constraint which makes it less robust to implement. In recent time, massive deployment of sensor based measuring devices into the network which acquires large amount of data for analysis. It facilitates data driven approaches over model-based approaches. This paper studies Data-driven ANN model for mapping of input parameters (PD, QD) and output parameters(V,) to emulate actual system. Additionally, three weight training algorithms namely (a) Levenberg-Marquardt (LM), (b) Bayesian Regularization (BR), and (c) Scaled Conjugate Gradient (SCG)are compared with MATPOWER embedded solver (MIPS) on the standard IEEE-33 bus distribution network. The LM, BR, and SCG based ANN models have the computational time of 0.009511,0.009129, and 0.011519 seconds respectively. It shows that the proposed BR-based models outperform rest of models, and also it is 254 time faster than the traditional MIPS solver.