New method to estimate soiling losses before PV project implementation

Scientists in Peru have proposed a self-contained, deployable system that quantifies energy losses from dust accumulation on PV modules. It uses both artificial neural networks and electrical models for soiling loss prediction.

Researchers from Peru have developed a self-contained, deployable system that can reportedly quantify energy losses from dust accumulation on PV arrays before their implementation.

The proposed approach combines methods from neural networks and incremental conductance, which is one of the most commonly employed maximum power point tracking (MPPT) techniques due to its simplicity and low implementation complexity.

“This system facilitates soiling analysis both before implementation and during the operation of PV systems. It can account for spatial inhomogeneities of dust accumulation and the long-term degradation of PV modules through recalibration and retraining,” the scientists explained.  “The proposed system integrates an artificial neural network (ANN), an electrical model, and MPPT based on incremental conductance. This combination allows for estimating daily energy losses due to soiling in a PV array with minimal maintenance requirements.”

The proposed method was tested on a system containing a 5W monocrystalline silicon PV test module, a DC-DC single-ended primary inductance converter (SEPIC), a pyranometer, module temperature sensors, and a computer with a 28 nm ARM Cortex-172 processor that runs the ANN. In addition to the ANN predictions, the group tested the system using an electrical model that uses iterative solving, meaning it repeats the calculation multiple times until it gets a good result. Both systems use measured solar irradiance and temperature as inputs.

The testing took place between September 2020 and September 2021, and the PV module was cleaned once a month. The data from the first month was used to train the ANN model, using 14,000 measurements in 150 iterations. The electrical model, on the other hand, did not need training as it is based only on the measured data and the PV specs.

“Both models exhibited comparable performances in estimating a clean PV module’s energy output, with the ANN model demonstrating lower computational costs,” they said. “The ANN model also showed slightly better accuracy, with a mean absolute percentage error (MAPE) of 0.5% compared to 0.6% for the electrical model. These results indicate that while both models are effective, the ANN model offers advantages in terms of computational efficiency and adaptability for retraining to compensate for long-term module degradation.”

The results of the 1-year testing showed that with a monthly cleaning schedule, energy losses due to soiling oscillated between 4% and 7% during most months. Elevated losses of up to 10% were recorded in months with nearby construction activities. “Our findings demonstrate the system’s capability to accurately predict performance losses due to soiling without the need for a complete PV system setup,” the researchers highlighted.

Concluding their work, they added that “the Incremental Neuroconductance system presents a robust and flexible solution for quantifying soiling losses in PV modules, contributing to more effective maintenance schedules and improved PV plant performance.”

The system was presented in “Incremental neuroconductance to analyze performance losses due to soiling in photovoltaic generators,” published in Energy Reports. The research was conducted by scientists from Peru’s National University of San Agustín and the Pontifical Catholic University of Peru.