Integrating multiple renewable energy sources and battery storage systems into modern power grids requires sophisticated control strategies to maintain stability and efficiency. Researchers have now advanced this capability through a novel machine learning-assisted control framework designed for multiport autonomous reconfigurable solar power plants—systems that combine solar generation, energy storage, and power conversion equipment connected to both alternating current and high-voltage direct current grid links.
The core challenge addressed involves capacitor voltage imbalances that develop when diverse power sources feed individual converter submodules through DC-DC converters. These imbalances constrain safe operating zones and reduce overall system efficiency. Traditional energy balancing control methods struggle to optimize performance across the wide range of real-world conditions these systems encounter.
The research team developed a hybrid approach combining conventional energy balancing control with machine learning algorithms that intelligently determine when to activate or deactivate balancing mechanisms. This reduces unnecessary control actions while maintaining voltage equilibrium across all submodules. The ML component learns operational patterns and predicts optimal control responses, effectively expanding the stable operating envelope of the system.
Validation occurred through two rigorous testing phases. Computer simulations using PSCAD/EMTDC software demonstrated theoretical feasibility, while hardware-in-the-loop testing—where physical control equipment interfaces with digital system models—confirmed practical performance under realistic conditions.
The significance extends beyond individual plant performance. By enabling multiport solar systems to operate safely and efficiently across broader conditions, this technology facilitates deeper renewable integration into existing grids. The approach supports grid modernization efforts worldwide, particularly as utilities deploy increasingly complex hybrid generation and storage facilities. The methodology also provides a template for similar machine learning applications in other advanced power electronics systems.
