Power Grid — Concepts & Terminology

1 · The fundamental rule: supply must equal demand

Unlike water or gas, electricity cannot be easily stored at grid scale. Every second, the total power generated must exactly match the total power consumed, plus transmission losses. If they drift apart even slightly, the grid frequency moves — and if it moves too far, equipment trips and blackouts cascade.

Balance equation

Gen = Load + Loss

Must hold every second

Nominal frequency

50 Hz

India / EU (60 Hz in US)

Safe band

49.9–50.05

IEGC operating band (Hz)

Trip threshold

±0.5 Hz

Under-frequency load shed

2 · Frequency is the heartbeat — and inertia is the flywheel

Grid frequency is a direct readout of the supply-demand balance. When demand exceeds supply, generators slow down and frequency falls. When supply exceeds demand, frequency rises. The rate at which it changes depends on system inertia.

df/dt = −ΔP / (2 · H · S)

Where ΔP is the power imbalance, H is system inertia constant (seconds), and S is system base power. More inertia = slower frequency change = more time to react.

Why inertia matters for renewables: Big spinning turbines (coal, gas, hydro) store rotational kinetic energy that naturally resists frequency change — like a heavy flywheel. Solar panels and wind inverters have no spinning mass, so as the grid adds renewables, total inertia drops, frequency changes faster, and balancing becomes harder. This is one of the central challenges AI helps solve.

Live frequency response calculator

A 1000 MW generator trips. See how fast frequency falls based on system inertia.

Lost generation (MW) 1000 MW

System inertia H (sec) H = 5.0

Rate of change (RoCoF)

−0.50 Hz/s

Frequency falling

Time to reach 49.5 Hz

1.0 s

Before reserves must act

3 · Reserves — the layered safety net

Operators hold spare capacity ready to respond at different speeds. Like the frequency response ladder, each layer buys time for the next.

Reserve type	Responds in	Source	Purpose
Inertial response	0–2 sec (instant)	Spinning mass	Slows the initial frequency fall
Primary (FCR)	2–30 sec	Governor droop	Arrests the fall, stabilises frequency
Secondary (AGC)	30 sec – 15 min	Automatic Gen Control	Restores frequency to 50 Hz
Tertiary	15 min – 1 hr	Manual / market	Replenishes used reserves

Battery storage changes the game: Grid-scale batteries respond in milliseconds — faster than any spinning generator — making them ideal for primary frequency response and synthetic inertia. This is why battery energy storage systems (BESS) are being deployed rapidly alongside renewables.

4 · Economic dispatch & merit order

Among all the generators that could supply power, which should run? Economic dispatch picks the cheapest combination that meets demand while respecting grid constraints. Generators are stacked in merit order — cheapest first.

Solar / Wind

~₹0/MWh

Nuclear

₹2,800

Hydro

₹3,200

Coal

₹4,500

Gas peaker

₹9,000+

The marginal price — what the most expensive running generator costs — sets the market clearing price that everyone is paid. This is why dispatching renewables first (near-zero cost) lowers prices for the whole grid.

5 · Why renewables make balancing harder

Challenge	What it means	AI / mitigation
Intermittency	Sun and wind vary minute to minute	ML forecasting of generation
Low inertia	Inverters add no spinning mass	Synthetic inertia from batteries
The duck curve	Steep evening ramp as solar sets	Storage + flexible demand
Forecast error	A cloud bank can drop 500 MW fast	Probabilistic forecasting + reserves
Curtailment	Too much solar at noon, nowhere to put it	Battery charging + demand shifting

6 · Where AI helps the most

In rough order of impact:

Renewable forecasting — ML predicts solar/wind output hours ahead from weather data, shrinking the reserves needed.
Demand forecasting — accurate load prediction means generators are scheduled efficiently, not wastefully.
Optimal dispatch (MPC/RL) — solving the economic dispatch problem in real time across thousands of constraints.
Battery optimisation — deciding when to charge/discharge for arbitrage and frequency response.
Contingency prediction — spotting equipment likely to fail before it trips and cascades.

Google DeepMind precedent: DeepMind's ML applied to wind farm output forecasting boosted the economic value of wind energy by roughly 20% by predicting output 36 hours ahead — letting operators commit to deliveries instead of holding expensive backup reserves. The same forecasting-plus-optimisation pattern underlies modern AI grid balancing.