In the heart of China’s Sichuan Province, a team of researchers from Sichuan University is tackling one of the energy sector’s most pressing challenges: how to optimize the scheduling of cascade hydropower and solar-powered reservoirs in a way that reduces costs, controls operational risks, and enhances renewable energy accommodation. Led by Dr. Chen Shi, the team has developed a novel approach that combines deep learning and reinforcement learning to address the complexities of high-penetration renewables and large-scale pumped storage.
The team’s research, published in the journal *Power Construction* (Dianli jianshe), focuses on the dual uncertainties of renewable energy output and interval water inflow, which can make system operational coordination increasingly complex. “Traditional scheduling methods often fall short when it comes to handling these uncertainties,” explains Dr. Chen Shi. “Our goal was to develop a method that could not only predict these uncertainties but also optimize scheduling strategies to mitigate risks and improve system flexibility.”
The team’s solution involves a multi-step process. First, they employ the “Informer” deep neural network to predict basin-interval water inflow, transforming inflow uncertainty into flexibility supply indicators. This step alone has shown significant improvements over conventional time-series approaches, with a 18.9% reduction in Mean Squared Error (MSE) and a 58.8% reduction in Mean Absolute Error (MAE). “The Informer model’s ability to capture peak events with a 78.5% accuracy rate is particularly noteworthy,” says Dr. Chen Shi.
Next, the team integrates risk theory to quantify flexibility deficits using the conditional-value-at-risk measure. This risk-aware approach allows for a more nuanced understanding of potential operational risks. Finally, they develop an improved risk-managing proximal policy optimization (RM-PPO) reinforcement-learning algorithm to derive optimized scheduling strategies.
The results of their research are promising. The RM-PPO scheduling algorithm achieves flexible system regulation through pumped-storage plants coordinated with photovoltaics (PVs) for surplus-energy absorption, while conventional hydropower stations synergize with PVs for source-load dynamic matching. This approach not only reduces costs but also controls operational risks while maintaining system flexibility.
The commercial implications of this research are significant. As the energy sector continues to shift towards renewable sources, the need for effective scheduling methods that can handle the uncertainties of renewable energy output and water inflow will only grow. Dr. Chen Shi’s team has provided a robust solution that could be applied to similar systems worldwide, enhancing renewable energy accommodation and improving hydropower utilization efficiency.
Looking ahead, this research could shape future developments in the field by promoting the integration of advanced machine learning techniques into energy management systems. As Dr. Chen Shi notes, “Our work is just the beginning. The potential for further advancements in this area is immense, and we are excited to see how our research can contribute to a more sustainable and efficient energy future.”
In a rapidly evolving energy landscape, the work of Dr. Chen Shi and his team offers a glimpse into the future of optimal scheduling for cascade hydropower and solar-powered reservoirs, providing a blueprint for the energy sector to follow.