Brazil’s distributed solar fleet achieves 72% average performance ratio

The average performance ratio (PR) of solar micro- and mini-generation systems in Brazil is 72%, with variations throughout the day and across different months due to factors such as seasonality and soiling. The estimate was presented at a technical workshop organized by the National Electric System Operator (ONS), in partnership with the German development agency GIZ and the Center for Research, Development, and Innovation Management (CGPDI), focused on improving generation forecasting methodologies for distributed micro- and mini-generation (MMGD) in Brazil.

In Brazil, microgeneration systems are defined as those with an installed capacity of up to 75 kW, while minigeneration systems range from more than 75 kW to 5 MW for renewable sources such as solar PV.

The study, presented by Lucas Nascimento, a professor at the Federal University of Santa Catarina (UFSC), analyzed data from nearly 8,000 MMGD plants selected from an initial sample of more than 9,000 installations provided through a partnership with SolarZ.

Nascimento explained that the 72% PR value is calculated based on direct current (DC) output. As Brazil’s National Electric Energy Agency (Aneel) database reports installed capacity in alternating current (AC), the ONS’s new solar generation forecasting methodology applied an average oversizing factor of 120%, resulting in an AC-equivalent PR of approximately 85%.

The rapid expansion of MMGD capacity in recent years has increased challenges for the National System Operator in monitoring net load data. This growth has contributed to steep ramps of up to 30 GW over a few hours and a significant reduction in minimum daytime load, a phenomenon known as the “duck curve” effect.

Even after adjustments for PR and irradiance, forecast load curves continue to diverge from observed data, which show a deeper daytime decline than expected. The study suggests that the remaining discrepancy is linked to the installed capacity variable, potentially due to delays in data updates or incomplete information in Aneel’s public database, which is sourced from distribution companies.

The machine learning models were validated through cross-validation using one year of data. The ONS plans to update its decision-making processes and implement a semi-hourly model for load forecasting.