Availability Estimation for Aggregated Battery Systems

Sammanfattning

Battery energy storage systems, also known as BESS, are increasingly used to provide frequency regulation services due to their fast response and operational flexibility. Accurate real‑time estimation of available power is crucial for balancing service providers to meet contractual commitments. This estimation depends on telemetry data sampled at finite temporal resolutions, but the impact of the sampling frequency on estimation accuracy has not been systematically quantified, especially for aggregated battery units.This thesis investigates how temporal resolution affects the estimated available power of an aggregated two‑unit BESS. A conceptual framework based on three physical constraints, namely the inverter power limit, the C‑rate limit, and the energy limit, is developed and implemented as a two‑stage software pipeline. The first stage continuously retrieves real‑time operational data from a live BESS installation via the Victron VRM API. The second stage computes available power for individual sites and for a virtual aggregated battery under different down‑sampling intervals ranging from 1 minute to 1 hour. This is done by retaining only the last State of Charge (SOC) value in each time window. Estimation errors are quantified using two metrics: the mean absolute error (MAE) and the maximum relative error (MRE).The evaluation uses one week of real data from two battery sites with distinctly different SOC dynamics. Results show that even at a one‑hour sampling interval, the MAE of the aggregated system and the less dynamic site remains below 1.03 × 10⁻³ kW, while the more dynamic site still exhibits a discharge MAE below 5.33 × 10⁻³ kW and a charge MRE below 5%. Aggregation of the two units consistently reduces estimation errors compared to the worst individual site. The findings indicate that one‑hour SOC sampling is sufficient for accurate available power estimation in typical stationary BESS applications with moderate SOC dynamics, offering a practical trade‑off between estimation accuracy and data acquisition overhead.