Non-Linear Temporal Machine Learning Models for Conditioning Monitoring in Large-Scale Solar Energy Systems

  • Maaløe, Lars (PhD Student)
  • Winther, Ole (Project Participant)

Project Details

Description

PhD project in cooperation with the Technical University of Denmark and GreenGo Energy. GreenGo Energy installs and operates photovoltaic systems for business, housing associations and public entities. The large number of condition monitoring sensors installed in all GreenGo Energy solar energy power plants generates terabyte data that are collected in a common cloud based solution. A global scale synchronized data acquisition system providing data with unprecedented precision, size, geographical diversity provides unique possibilities for big data modeling. A successful machine learning system build on top of the cloud solution will be able to detect many types of faults and wear characteristics. The scale of the data poses computational challenges and requires application and development of novel non-linear dynamical models that scale to large datasets. In the PhD project, Bayesian approaches to filtering will be investigated as well as deep learning methodologies for integration of high frequency heterogeneous sensor data. The service platform will integrate state-of-the-art fault diagnosis, and portfolio based service planning and execution automation.
StatusFinished
Effective start/end date15/12/201415/12/2017