Soiling Rate Determination from Referenced Systems in Desert Climate using PVInsight Soiling Algorithm

We present a comparison of two methods to estimate the soiling losses from photovoltaic (PV) system operational data. We evaluate 21 months of data from an outdoor soiling test in a hot, desert climate with a reference system which was cleaned once a week. First, we present an analyst-based result that uses accepted industry practices to estimate trends based on data fit and records of cleaning and storm events, representative of an industry-standard, labor-intensive approach. Second, we present a completely automated analysis, in which the cleaning events are selected through an unsupervised statistical learning algorithm. The algorithm generates estimates of linear soiling rates that are in close agreement with the rates determined by the human analyst, at orders of magnitude faster speed. The speed of processing is particularly advantageous in analysis of multiple setups and long duration data (years of minute(s) data collection).