An ANN-based distributed secondary controller used to ensure accurate current sharing in DC microgrid

Abstract

In a DC Microgrid, to minimise the effect of unequal cable resistances, distributed secondary controllers are included in addition to droop controllers. An accurate Current Sharing based Secondary Controller (ACSSC) is proposed in this paper, which is used to ensure proportional or accurate sharing of power demanded by loads among sources in a DC microgrid. However, ACSSC is a model approach and imposes a computational burden on digital signal processors (DSPs) in the case of a DC microgrid having a large number of sources and loads. To address this issue, a distributed Artificial Neural Network-based ACSSC (ANN-ACSSC) is proposed, which includes a learning framework of the accurate current estimator, droop gain estimator and voltage estimator to ensure accurate current sharing in the DC microgrid and maintain the voltage regulation across the converter within the specified limit. The performance of ANN-based estimators is compared to the state-of-the-art machine learning-based secondary controllers like bagged ensemble, Reinforcement-learning-based Integrated Control (RLIC) and Reinforcement Learning-based Approximate Dynamic Programming (RLADP-based estimators using performance metrics like MAE, MSE and R² score. The inference time of various estimators included in ACSSC is evaluated using the ‘tic-toc’ functionality included in Matlab. The Levenberg-Marquardt algorithm is used for training the various estimators included in ANN-ACSSC. The performance of the proposed ANN-ACSSC is compared with that of the proposed ACSSC. Further, the performances of ANN-ACSSC and ACSSC for a nonzero value of communication delay are studied. The viability of the ANN-ACSSC is validated using the experimental results captured using a lab prototype of a DC microgrid.