In the ever-evolving landscape of renewable energy, the integration of photovoltaic (PV) systems into power grids has become a double-edged sword. While these systems help reduce carbon footprints, they also introduce significant challenges in maintaining stable voltage levels. Traditional methods, reliant on devices like Load Ratio Control Transformers (LRTs) and Static Capacitors (SCs), struggle to keep up with the rapid fluctuations caused by weather-dependent PV outputs.
Enter Fumiya Matsushima, a researcher from the Department of Electrical and Mechanical Engineering at Nagoya Institute of Technology, who has developed a groundbreaking solution. Matsushima’s work, published in ‘Energies’ (Energies), introduces a Machine Learning (ML)-based control method that could revolutionize voltage management in sub-transmission grids. His approach combines long-term LRT tap-changing with short-term reactive power control of Power Conditioning Systems (PCSs), using Deep Reinforcement Learning (DRL) to optimize both.
The crux of Matsushima’s method lies in its ability to estimate voltages at each grid node using a Deep Neural Network (DNN) that processes measurable substation data. “By using limited measurable information, we can achieve real-time voltage monitoring and control, even in grids without extensive sensor installations,” Matsushima explains. This is a game-changer, as it enables comprehensive grid-wide voltage control without the need for costly and extensive sensor installations.
The DNN estimates voltages based on real-time data from substations, which are then used by DRL agents to determine optimal LRT tap positions and PCS reactive power outputs. This multi-timescale control strategy ensures that both long-term and short-term voltage fluctuations are effectively managed.
But Matsushima didn’t stop at developing a highly effective control method. He also addressed the “black box” nature of DRL models, which often makes their decision-making processes opaque. By applying Shapley Additive Explanation (SHAP), an Explainable AI (XAI) technique, Matsushima made the DRL model’s decision-making process more transparent. “SHAP enhances interpretability and confirms the effectiveness of the proposed method,” Matsushima notes. This transparency is crucial for the adoption of DRL in the energy sector, as it allows engineers to understand and trust the model’s decisions.
The implications of this research are vast. For the energy sector, it means more stable and efficient power grids, which can accommodate a higher penetration of renewable energy sources without compromising reliability. This could lead to significant cost savings and improved grid performance, benefiting both utilities and consumers.
As the energy sector continues to evolve, the need for intelligent and adaptive control systems will only grow. Matsushima’s work lays a strong foundation for future developments in this field. By combining ML-based voltage estimation with DRL for control, and incorporating XAI for transparency, Matsushima has set a new standard for voltage management in modern power grids. His research, published in ‘Energies’, is a significant step forward in the quest for more efficient and reliable energy systems.
