Classification of Failures in Photovoltaic Systems using Data Mining Techniques

Introduction

During recent years the number of large-scale PV (photovoltaic) systems has grown worldwide. In 2010, the photovoltaic industry production more than doubled and reached a worldwide production volume of 23.5 GWp of photovoltaic modules. Business analysts predict that investments in PV technology could double from € 35-40 billion in 2010 to over € 70 billion in 2015, while prices for consumers are continuously decreasing at the same time. (European Commission, DG Joint Research Centre, Institute for Energy, Renewable Energy Unit, 2011).

The complexity in PV system configurations represents an additional problem in maintenance and control operations in large systems. For example, a failure in one PV module placed at a big façade is very difficult to detect. A quick detection of failures would avoid energy losses due to malfunctions of PV system and therefore improve its performance and end-user satisfaction (Roman, Alonso, Ibanez, Elorduizapatarietxe, & Goitia, 2006).

When data-mining techniques are applied to a PV database, a wide variety of relations between parameters can be found. This study relays on expert knowledge to study the possible behaviours of PV generation performance that can be affected by changes in the environment conditions; furthermore, computing learning algorithms allow us to detect and classify failures without measuring the environmental parameters.

This study focuses on a methodology to control the correct performance of each group of modules which compose a large-scale PV generator, identifying failure occurrences and its most likely causes, and it does so by a procedure that analyses the performance of a group of modules by ignoring environmental information.