A public test of Laya on a home energy system: fast enough, but a tariff EMS still wins

# A test of Laya on a house

Laya does not write a paragraph. You give it the situation and the answers it is allowed to pick. It returns the pick and a probability, in one pass, on the machine in the house.

We wanted to know if that is useful in a home energy system. The ordinary job is when to charge the car, when to use the battery, and how to stay under the fuse. The job we hoped for is the messy minute: a charger that vanishes, a meter that disagrees with the devices, a register that freezes while the sun moves. Those are the cases that turn into a long list of special rules.

This repository is that test. One simulated Swedish villa. Four controllers on the same minutes. The record of the day is below, and the lab notes are in [docs/EMS_FEASIBILITY.md](docs/EMS_FEASIBILITY.md).

## What happened

| Controller | Bill | 15-min peak | Car at 07:00 | Decision time |
|---|---:|---:|---:|---:|
| Naive charger | 985 SEK | 11.04 kW | 100% | under 1 ms |
| Laya, then a safety clip | 530 SEK | 6.25 kW | 100% | p95 179 ms |
| Tariff EMS | 385 SEK | 4.66 kW | 80% | under 1 ms |
| Mathematical plan | 258 SEK | 2.95 kW | 80% | 158 ms, once |

The tariff EMS is the better controller. Laya is fast enough for a one-minute loop, and the fuse holds. It is useful against a charger that has no plan. It is not a replacement for a short program that knows the prices, and it did not read the fuzzy faults well enough to trust it with them.

## The house

A 16 A villa. The fuse is 11.04 kW. On the roof, 10 kWp of PV. In the garage, a 10 kWh battery that can move 5 kW, and an EV with a 60 kWh pack and an 11 kW charger. The car comes home at 17:30 with 35% charge and should leave at 07:00 with 80%. Export is paid at 80% of the spot price. The grid fee we care about is 81 SEK for each kW of