Fault Detection and performance visualisation for a grid-connected Photovoltaic Power Plant using sensor data

Dyamond, Wayne Peter (2019-12)

Thesis (MEng)--Stellenbosch University, 2019.

Thesis

ENGLISH ABSTRACT: The rising energy demand and need for alternatives to fossil fuel based power generation have increased the utilisation of photovoltaic (PV) power plants. The reliable operation of PV power plants will maximise energy delivery, boost public opinion on PV technology and lead to financial gains for investors. Accurate fault detection and effective plant performance reporting could significantly reduce system downtime, power loss and safety hazards. The work presented in this document aims to investigate improvements to fault detection and performance visualisation for an utility-scale PV power plant using measured sensor data. 560 GB of operational data from a 75 MWp capacity solar power plant is obtained for the research project. Data pre-processing and cleaning results in a 167 GB dataset containing measured values for 12 595 different signals over the period of three years. A fault detection procedure based on the comparison of modelled and measured string-pair current is proposed. The expected current is modelled using the single diode electrical model. The Euclidean distance between the measured and expected values is calculated for all string-pairs in the power plant. Events are flagged as possible faults when the corresponding Euclidean distance is considered an outlier. The fault detection procedure is tested on the dataset and a sample accuracy of 94:67% is achieved. A visualisation tool based on the performance comparison of all string-pairs is developed. The visualisation is used to verify events detected during the fault detection procedure as well as visualise average performance and degradation differences between string-pairs. An average DC degradation rate of 0:38% per year is observed during string-pair degradation analysis.

AFRIKAANSE OPSOMMING: Die toenemende energie aanvraag en behoefte om alternatiewe bronne vir energieopwekking te gebruik, het gelei tot die ontwikkeling van meer fotovoltaïese (FV) kragstasies. Betroubare werksverrigting van FV-kragsentrales sal die energie-opbrengs verhoog, die publiek se mening oor FV-tegnologie verbeter en tot hoër winste vir beleggers lei. Akkurate foutopsporing en effektiewe verslagewing van aanleg werksverrigting kan stelsel stiltand, kragverlies en veiligheidsrisiko's aansienlik verminder. Die navorsing wat in hierdie dokument uitgel ê word, mik om verbeteringe aan foutopsporing en visualisering van stelsel uitset vir 'n netwerkverbindte FV-kragstasie te ondersoek. 560 GB se gemete sensor data van 'n 75 MWp sonkragaanleg word in hierdie navorsingsprojek ondersoek. Verwerking van die data verminder die grote tot 167 GB. Die datastel bevat meetingswaardes vir 12 595 verskillende bronne vir drie jaar. 'n Foutopsporingprosedure, gebaseer op die vergelyking van gemodelleerde en gemete stringpaarstroom waardes, word aangebied. Die verwagte stroom word gemodelleer met behulp van die enkeldiode elektriese model. Die Euklidiese verskil tussen die gemete en verwagte waardes word bereken vir alle stringpare in die kragsentrale. Uitskieters word as moontlike foute geïdentifiseer. Die foutopspooringprosedure word op die datastel getoets en behaal 'n steekproef akkuraatheid van 94:67%. 'n Visualiseringstoepassing, wat die werkverrigting van alle stringpare vergelyk, word ontwikkel. Die visualisering word gebruik om gebeurtenisse wat tydens die foutopspooringprosedure geïdentifiseer is, te bevestig. Die toepassing word ook gebruik om die verskille in gemiddelde uitset en agteruitgang van drywing tussen stringpare te visualiseer. 'n Gemiddelde gelykstroom-agteruitgangskoers van 0:38% per jaar word waargeneem tydens die analise.

Please refer to this item in SUNScholar by using the following persistent URL: http://hdl.handle.net/10019.1/107181
This item appears in the following collections: