Dalarna University's logo and link to the university's website

In recent year there has been a massive increase in photovoltaic (PV) installations in Sweden, with a 93% increase from the year 2019 to 2023. Even with all the advantages that PV electricity generation offers, it is important to understand the hurdles that come with its implementation, one of them being the degradation of PV modules. Understanding this degradation is integral so that stakeholders understand the financial implications of their investment, and for grid operators to plan accordingly the future of the grid. 

A problem arises when degradation is to be determined since traditionally a system needs to stop operating to be able to determine the degradation. Some researchers have tried tackling this problem by calculating the degradation rate through monitoring data by comparing different performance metrics over several years of operation, and the knowledge that degradation tends to be linear in most PV technologies. Despite several studies proving the usefulness of these methods, no study has been done in higher latitudes where seasonality complicates these calculations due to higher variations in the performance metrics utilized.

This study aims to assess the efficacy of the Linear Regression (LR), Year-on-year (YoY), and PVUSA characterization LR methods to calculate the degradation rate of a PV system in higher latitudes with different data aggregation windows. Furthermore, the thesis also aims to assess if it is possible to utilize data from weather stations to calculate the degradation rate, so systems without dedicated temperature, wind, and irradiation sensors can also quantify the health of their system without the extra capital investment that comes with the implementation of a measuring system.

The process of calculating the degradation starts with the identification and removal of faulty data.  At the same time temperature, wind speed, and irradiance data from nearby weather stations is extracted to utilize in the process. The irradiance data is decomposed through the Engerer 2 model and transposed with the Perez 1990 model with the help of Python code. With the cleaned and replaced data, then the performance metric is calculated, filtered and the degradation rate is calculated.

The results obtained for the scenario with no data replacement are -0.388 %/year for the PVUSA model, -0.676 %/year for the LR with monthly aggregation, -0.498 %/year for LR with weekly aggregation, -0.341 %/year for the YoY method with monthly aggregation, and -0.503 %/year for the YoY method with weekly aggregation.  

For the scenario where data was replaced, a degradation rate of -0.476 %/year for the LR with monthly aggregation, -0.393 %/year for LR with weekly aggregation, -0.085 %/year for the YoY method with monthly aggregation, and 0.0.22 %/year for the YoY method with weekly aggregation. No useful result for the PVUSA method was obtained when data was replaced.

Out of the Linear regression results, only the result from the monthly aggregated data where data was replaced (-0.476 %/year), and the result from the weekly aggregated data where there was no data replacement (-0.498 %/year) were statistically significant.

Results from the YoY method are statistically significant, but with a larger confidence interval value.