GENERAL, ROBUST AND SCALABLE METHODS FOR STRING LEVEL MONITORING IN UTILITY SCALE PV SYSTEMS

摘要

Due to the rapid growth in the size and number of photovoltaic (PV) power plants, monitoring solutions that ensure a high, reliable power output and low maintenance costs are of increasing importance. Three criteria for such monitoring solutions are generality (applicability to different data sets), robustness (ability to detect faults without producing false alarms) and scalability (numerical efficiency). Using data from three MW-scale PV plants located in Sub-Saharan Africa, the Middle East and Northern Europe, a simple fault detection algorithm has been proposed in light of these criteria. The algorithm uses production data as input, filters out unwanted datapoints and calculates a performance metric with a pre-defined frequency, before using this metric to evaluate the performance of the different sub-arrays. A main contribution of this work is the proposal of a procedure for selecting filtering thresholds to reduce noise in the performance metric. We show that by applying suitable filters, the sensitivity of the fault detection algorithm is increased 2 - 5 times, greatly improving the robustness of the algorithm.