Team: Dr. Selva Nadarajah, Arman Aminipanah
Overview
The Smart Grid Optimization project develops AI-powered decision support systems for real-time power grid management and load balancing. As the energy landscape shifts toward decentralized renewable sources, grid operators face unprecedented complexity in maintaining stability and efficiency. Our system uses reinforcement learning and stochastic optimization to make intelligent dispatch decisions that reduce costs while maintaining reliability.
Approach
We combine Markov decision process (MDP) models with deep reinforcement learning to create agents that can learn optimal dispatch policies in real time. The system ingests live sensor data from grid infrastructure, weather forecasts, and market prices to generate actionable recommendations for grid operators. Our self-adapting approximation algorithms allow the system to scale to grids with thousands of nodes without sacrificing solution quality.
Impact
Early simulations show a 12-18% reduction in operating costs and a 25% improvement in renewable energy utilization compared to traditional rule-based dispatch systems. We are working with regional grid operators to pilot the system in a real-world setting.
Related Publications
Physical vs. virtual corporate power purchase agreements: Meeting renewable targets amid demand and price uncertainty
D. Mohseni Taheri, S. Nadarajah, and A. Trivella
European Journal of Operational Research
Hierarchical planning for hydropower capacity upgrade: Exploiting structure in reoptimization and investment policies
A. Kleiven, S. Nadarajah, and S.E. Fleten
Working Paper
Comparison of least squares Monte Carlo methods with applications to energy real options
S. Nadarajah and N. Secomandi
European Journal of Operational Research
Merchant energy trading in a network
S. Nadarajah, F. Margot, and N. Secomandi
Operations Research