5. 3 Operation, Performance and Maintenance of PV Systems
PV System, PV System Performance, Data Imputation, Data Quality
Summary / Abstract:
In the framework of the new H2020 project TRUST PV, the Cost Priority Number methodology, a cost-based failure modes & effect analysis method, is tailored for its application onto real-case studies during the operational phase of a PV plant. Thereby, it is possible to calculate the energetic and economic losses for each individual failure by combining actual technical characteristics of the failure tickets, the operational costs of fixing those failures, and the power loss due to the failure occurrence. In this work, different empirical, machine learning and univariate models have been tested to find the optimal model to impute power data for all possible scenarios of missing performance data. The parameter space comprises of the amount and type of missing data, the ratio between training and test set and the predictor availability. Overall, the most desirable setting for highly accurate data imputation is the availability of a neighbouring string/plant or measured climate data as predictor with a high training/test set ratio for small test sets and ratios between 80/20 to 50/50 for bigger test sets (longer than 12 hours). In general, machine learning imputation models and empirical power models perform similar for smaller test set sizes and machine learning models are preferrable for bigger data gaps. Univariate methods should be avoided if possible.