Artículos de investigación

PVRW

marzo 6, 2026

Integrated Performance Diagnosis of a Utility-Scale Photovoltaic Plant

Por:

Rafaél Ávila Naranjo

Introduction

Utility-scale photovoltaic (PV) plants frequently exhibit discrepancies between measured energy production and model-based expectations, leading to uncertainties in performance assessment, operational diagnostics, and financial forecasting.

A 9.8 MWac PV plant in Colombia showed persistent underperformance, reflected in a Performance Ratio (PR) below target and an Energy Performance Index (EPI) systematically different from unity. These deviations may arise from both technical faults and modeling inaccuracies.

This work presents an integrated performance diagnosis framework that combines:

multi-year operational data quality control,
machine learning–based fault detection, and
physically updated PV simulation modeling,

to distinguish real operational losses from modeling bias and improve the accuracy of energy predictions.

Methodology

An integrated diagnostic framework was developed combining data analytics, automated fault detection, and physics-based modeling:

Data quality control: cleaning and validation of meteorological, inverter, and plant-level meter data.
Performance assessment: evaluation of PR, EPI, and maximum power behavior.
Machine learning: anomaly detection and fault classification at inverter and plant levels.
PVsyst modeling: updated simulation including real shading, degradation, and operational losses.
Validation: measured vs. simulated comparison to improve energy prediction accuracy.

Results

Stage 1 – Data Quality and Energy Performance:

A comprehensive assessment of operational data quality and performance indicators was conducted to establish a reliable baseline of the plant’s actual behavior and to quantify deviations between measured and expected energy production.

Stage 2 – Fault Detection and Energy Impact

Using the cleaned datasets, automated anomaly detection techniques were applied to identify operational faults at both inverter and plant levels, and to estimate their direct contribution to annual energy losses.

Inverter Power Limitations
Persistent clipping of inverter output below expected power.
Occurred under normal irradiance conditions.
Not associated with grid curtailment.

Inverter Shutdowns
Sudden drops to zero production while irradiance remained stable.
Recurrent events affecting specific inverters.
Indicative of protection trips, communication issues, or electrical faults.

Plant-Level Failures
Simultaneous generation loss across the entire plant.
Abrupt fluctuations or complete outages.
Associated with operational or interconnection events.

Stage 3 – Modeling and Validation

Finally, the simulation model was adjusted by incorporating site-specific physical conditions and validated against historical measurements to reduce prediction bias and improve energy estimation accuracy.

Conclusions

The PV plant exhibits significant data quality issues, with up to 35% missing data in the main energy meter and extended timestamp gaps, highlighting the need for improvements in data acquisition and monitoring systems.
Abnormal inverter behaviors were identified, mainly power limitations and spontaneous shutdowns, directly impacting energy production and requiring detailed on-site diagnostics for effective mitigation.
The ICREA diagnostic framework enables the automatic detection and classification of subtle faults that are not evident through conventional performance metrics, supporting preventive and corrective maintenance strategies.
The updated PVsyst simulation model, incorporating real site conditions and operational losses, significantly improved the representation of plant behavior, reducing prediction bias and providing more reliable energy estimates.

References

[1] AbdukMawjood, K., Shady, S., & Walid, G. (2018). Detection and prediction of faults in photovoltaic arrays: A review. Proc. IEEE CPE-POWERENG.

[2] China General Certification Center & Huawei Technologies Co. (2020). Smart I-V Curve Diagnosis. Technical report.

[3] Jordan, D., & Hansen, C. (2023). Clear-sky detection for PV degradation analysis using multiple regression. Renewable Energy.

[4] Kontges, M., Kurtz, S., Packard, C., Jahn, U., Berger, K., Kato, K., & Friesen, T. (2014). Review of Failures of Photovoltaic Modules. IEA-PVPS T13-01.

[5] Parson, C. (2021). What is a machine learning model? NVIDIA Technical Blog.

[6] Reno, M., & Hansen, C. (2016). Identification of periods of clear-sky irradiance in GHI time series. Renewable Energy.

[7] Venkatakrishnan, G. R., et al. (2013). Detection, location, and diagnosis of faults in large PV systems: A review. Int. Journal of Low-Carbon Technologies, 659–674.

Brindamos soluciones integrales para el desarrollo de proyectos fotovoltaicos en Colombia y Latinoamérica

Menú

Contáctanos

× ¿Cómo puedo ayudarte?