ML-Driven Distribution Network Aggregation Considering Load and Inverter-Based Resources

The integration of distributed generation sources, such as photovoltaic (PV) systems and battery energy storage systems (BESS), is revolutionizing traditional power distribution networks, presenting new challenges in grid management due to the dynamic and bidirectional flow of electricity. Accurate estimation of power system parameters is essential for optimizing performance and maintaining grid stability; however, existing methods often suffer from inadequate datasets that fail to reflect the complexities introduced by these advanced technologies. This research addresses this gap by leveraging modern machine learning (ML) techniques to enhance the accuracy of parameter estimation for network aggregation in active distribution networks.Using the IEEE 33-bus system and incorporating control modes as defined under the IEEE 1547 standard, we developed a robust data generation framework. This framework utilizes OpenDSS simulations and automated data extraction methods to create comprehensive and diverse datasets, represented as rank-3 tensors of inhomogeneous shape. These datasets were employed to train ML models capable of predicting key parameters, including voltage per unit, active power loss, reactive power loss, new ZIPV parameters, and power bases across various operational scenarios. Our findings demonstrate that the ML models achieved a mean squared error (MSE) as low as 0.015 for the loadsonly model and 0.014 for the PV-BESS system, highlighting the potential for high prediction accuracy in real-world applications.