Energy Grid → Question 8

Does Battery Intelligence Matter?

The heuristic is simple: discharge when there's a deficit, charge when there's surplus. Dynamic programming finds the theoretical best. The gap between them tells you how much value you're leaving on the table — and how good your forecast needs to be.

Question 8 — The Gap

The Gap

The heuristic battery dispatch rule is deliberately simple: charge when state-of-charge drops below 40%, discharge when there's a deficit. It works — but how much better could you do? Backward dynamic programming with perfect foresight over the full year provides the theoretical upper bound. The gap between them is the value of intelligence.

Each configuration below shows heuristic vs. optimal unserved energy. The bigger the gap, the more a smarter dispatch policy could help — without adding a single megawatt of capacity.

Unserved Energy: Heuristic vs. Optimal Dispatch (TWh)
Finding
Perfect-foresight dispatch reduces unserved energy by 27–98% depending on the configuration. The benefit is largest where margins are tightest — exactly where reliability decisions are made.
Question 8 — The Key Result

The Forecast Question

Perfect foresight is a theoretical construct — nobody knows the full year of weather ahead of time. The real question is: how much forecast do you actually need? Receding-horizon optimization with limited lookahead reveals the answer.

For the 70 GW gas + 120 GWh storage configuration, here's what happens as you extend the forecast window from zero (heuristic) through 6, 12, 24, and 48 hours, all the way to perfect foresight:

Unserved Energy by Forecast Horizon (70 GW + 120 GWh)
Value Captured with 6-Hour Forecast
97%
of perfect-foresight dispatch value, using only a 6-hour weather forecast that ERCOT already produces.
Finding
A 6-hour weather forecast captures 97% of the perfect-foresight dispatch value. ERCOT already has 6-hour-ahead wind forecasts. The technology exists — the optimization just needs to use it.

Model: backward dynamic programming (perfect foresight) and receding-horizon DP (limited forecast). Hourly resolution, 8,760 timesteps. 70% RE scenario, 15 GW DC solar. Battery: 90% round-trip efficiency, 1C charge/discharge rate. Heuristic: charge from gas when SOC < 40%, discharge when demand exceeds supply.

Question 8 — Implications

What This Means

This is NOT an AI finding

The optimization is dynamic programming, not machine learning. No training data, no neural networks, no gradient descent. It's classical operations research applied to battery dispatch — a technique from the 1950s. The value comes from the math, not the buzzword.

The heuristic's blind spot

The heuristic charges from gas only below 40% SOC. The optimizer learns to pre-charge before forecast wind drops and hold reserve for evening peaks. These are simple rules that could be implemented without DP — the optimization just reveals what the rules should be.

The fidelity lesson

Dispatch policy fidelity matters as much as capacity fidelity. Smarter dispatch doesn't replace gas capacity, but it wrings more value from the storage you already have. Getting the "how much to build" question right is half the job. The other half is operating it well.

← Prev Q7: Cascade Study Back to Energy Grid