Fit the battery's conversion

The battery's efficiencies decide what a kWh through it costs (What a stored kWh is worth). A datasheet's round-trip figure is a start; the plant's own meters say what it really does, and how its loss grows with power. microgrid fit-battery fits the battery's parameters for plant/2 from a dataset of its meters.

What it needs

A dataset with, over the window (30 days by default):

• the battery's meter (battery, at its DC side, positive charging) and the SOC;
• the battery inverter's AC power (inverter, positive out to the house);
• the pv and grid meters and the house load;
• a DC-bus load's meter, if the plant has one (slots where it drew are left out).

Run it

uv run microgrid fit-battery --data my-dataset --end 2026-09-30 --days 30

--end is the local day the window ends at; it is required, since a dataset has no "now".

What it fits, and how

Two least-squares fits over 10-minute slot means, bootstrapped together by day for their intervals:

1. The inverter's loss. What goes in less what comes out, in every slot:

   loss = standby + working × (battery working) + l_c P_c + q_c P_c² + l_d P_d + q_d P_d²

   with P_c and P_d the battery's DC charge and discharge. The battery is "working" above working_threshold_kw (0.05 kW) either way. A PV term is fitted too, as a check that the PV meter reads AC: it should come out 0.

2. The battery's meter against its SOC. Each hour's change in stored energy (SOC × capacity) against its metered charge and discharge, without an intercept: stored = a × charged − b × discharged. By the hour, not by the day: a day charges about what it discharges, so a daily fit cannot tell a from b, while most hours only charge or only discharge.

It then prints the parameters for plant/2:

plant/2 field	From the fit
battery.eta_charge	a / (1 + l_c): AC to stored, at low power
battery.eta_discharge	(1 − l_d) / b: stored to AC, at low power
battery.loss_charge_per_kw	q_c
battery.loss_discharge_per_kw	q_d
battery.meter_charge	a
battery.meter_discharge	b
standby_kw, working_kw (the inverter's)	standby, working

each with its 90 % interval (sampling error only), the fit's residual by power band (flat if the form is right), and the same figures under simpler forms. The spread between forms is the larger uncertainty.

Read it

=> the plant's parameters (plant/2):
   [battery] eta_charge               0.94718 (90 % +0.9439 .. +0.9506)
   [battery] eta_discharge            0.95865 (90 % +0.9496 .. +0.9716)
   [battery] loss_charge_per_kw       0.00247 (90 % +0.0020 .. +0.0030)
   ...
   [hybrid] standby_kw                0.03021 (90 % +0.0300 .. +0.0306)
   [hybrid] working_kw                0.05547 (90 % +0.0523 .. +0.0600)
   ...
   discharging 1 kW DC delivers 86.6% of it, draws included

Copy the figures into your plant/2 and run check-plant to confirm the model's SOC tracks the meters better than before.

The fit assumes everything the battery stores or gives passes one converter that meters both of its sides (AC-coupled PV, the battery's inverter metering its AC power and the battery's DC). On other topologies, read the fit with care.