Optimum Power Flow in DC Microgrid Employing Bayesian Regularized Deep Neural Network

•This research paper proposes a comprehensive deep learning framework for power flow control in DC Microgrid.•The problem is formulated as intermittent operation of connected renewable energy sources reduces the reliability of power flow.•A Bayesian regularized Deep Neural Network approach is proposed to solve the problem.•The IEEE 9-bus test systems are utilized to perform the numerical analysis.•The simulation results and real-time experiment validate the performance and importance of the proposed model. Renewable Energy Sources (RES) in DC microgrid exhibit intermittent operation. This stochastic operation of RES affects DC Microgrid (DCMG). Unavoidable stochastic operation of RES makes supply-load mismatch. To provide compensation for supply-load mismatch, branches of DCMG have Battery Energy Storage Systems (BESSs) installed. Even though BESS are installed, their State of Charge (SOC) level needs accurate maintenance for balanced power flow. By employing Bayesian regularized Deep Neural Network (BDNN) a unique charging/discharging of BESS was achieved. The voltage quality and various reliability indexes were assessed to impact the validation of the proposed BDNN power flow control technique. The simulation results of the IEEE-9 bus system and real-experimental validation using EcoSense 2KW DCMG hardware in loop were conducted to validate the performance. The proposed BDNN power flow technique could reduce the Loss in Load Expectation (LOLE) from 2.1×10−4 to 7.8×10−8 without a change in Loss of Supply Expectation (LOSE) value. Also, our analysis for enhanced reliability shows that the Expectation Rise in Voltage (ERV) index increased from 3.5×10−5 to 4.6×10−2 for various PV penetration levels. Hence, conducted case study results confirm the reliable operation of DCMG to meet the supply-load compensation during changes caused by intermittent RESs. Graphical Abstract [Display omitted] .