The rise of renewable generation, batteries, and distributed‑energy resources (DERs) has hit two big road‑blocks in realtime power system and electricity market operations. First, utilities can't easily predict how households will react when electricity prices change—especially during rare events like sudden outages—because of the heterogeneity of energy customers. Second, today's real‑time pricing is highly volatile: the price attached to electricity (the locational marginal price, or LMP) can soar to extreme highs—or even dip below zero—forcing generators to receive out‑of‑market "uplift" payments that distort incentives and add hidden costs.
At CERAWeek, Assistant Professor of Engineering Cong Chen showed how AI agents can simulate energy customer behaviors and how to design an LMP adder for real-time pricing. Her team built AI agents powered by large‑language models and assigned them distinct personalities (e.g., a data‑savvy PhD student, a cautious grandmother, and an emotionally-driven actor). By feeding each agent the daily electricity price and asking whether to charge or discharge a home battery, the agents responded in line with their personalities. After a simulated blackout, the PhD‑type continued selling power for profit, however, the grandmother saved energy for emergencies, and the actor also saved extra backup power. These virtual "people" expose the wide range of real‑world preferences that shape household decisions.
Because the agents run quickly and can be recreated in endless variations, they can give utilities a low‑cost test‑bed for "behavior discovery." Chen's experiments showed that in a simple market, customers tend to act predictably and near‑optimally; in a more complex or volatile market, their choices diverge, making grid planning tougher. Importantly, utilities can use this methodology to study how different personalities would behave during rare, high‑impact events—such as a blackout—without waiting for those events to happen in real life.
Chen also tackled the realtime pricing puzzle. The current LMP system can swing dramatically, sometimes turning negative, which forces generators and batteries to receive extra out‑of‑market "uplift" payments. She proposed two new uniform pricing formulas that keep all payments inside the market, remove negative prices, and support market transparency. Max Dispatch Cost Pricing (MDCP) adds a modest surcharge to the LMP while maintaining a low energy burden for customers. Max Temporal Locational Marginal Pricing (MTLMP) adds a "social" price adder to rebate electricity providers with their highest marginal benefit contributed to the power systems. Both approaches eliminate costly uplift subsidies and encourage generators to report their true costs.
Bottom line: AI agents that mimic the diverse personalities of real households can reliably forecast how people will respond to electricity prices, even during rare shocks. Understanding this behavioral richness lets policymakers design simpler, fairer market rules, while the new realtime pricing formulas remove hidden costs and perverse incentives in the electricity market. By pairing human‑like AI simulations with smarter price signals, we can move toward a power system that's both more responsive to everyday users and sturdier when the unexpected strikes.
CERAWeek is regarded as one of the most influential annual conferences in the energy sector, drawing more than 11,000 energy leaders, researchers, technologists, and entrepreneurs from around the globe.
At the 2026 conference, Dartmouth faculty members appeared in panels spanning organizational transformation, industrial data infrastructure, grid modernization, electricity market design, and wildfire risk management.
Explore Dartmouth's engagement at CERAWeek 2026.