Advanced modeling and simulation for large-scale predictive analytics

The project aims to realize a predictive analytics solution that allows to predict imminent failures and to schedule pro-active maintenance, demonstrated on the case of solar panels.

Summary of the project

The present proposal builds upon the emerging Internet-of-Things trend, that allows a company

  • to use real-time data from sensors to monitor remote machines and infrastructures,
  • to guide maintenance,
  • to optimize production processes,
  • to maximize yield.

The proposal aims to realize a predictive analytics solution that allows to estimate future performance, to predict imminent failures and to schedule pro-active maintenance based on

  1. advanced analysis of historical data by means of simulation and modeling techniques (e.g. advanced statistics, probabilistic and machine learning models), and
  2. combining the models derived from historical data with real-time data through a scalable complex event/stream processing infrastructure.

The proposed solution will be validated on energy-related production (log) data obtained from solar plants, provided by the industrial partner.

This project embrace two major project innovation:

  1. we will investigate whether the use of semantic technology, such as ontologies, can provide an open and evolvable solution for the lack of harmonization and standardization in the log file of the PV domain.
  2. different algorithms and methodologies for building suitable simulation and prediction models from the data need to be researched, improved or newly developed and evaluated to determine which one allow to apply predictive maintenance in real life industrial contexts.

This proposal takes place in the context of doctiris which aims to encourage partnerships between the academic sector and the industrial sector in the Brussels-Capital Region. Through a call for projects, the programme aims to finance (4 years) doctoral thesis projects in applied research carried out in collaboration with an industrial partner (in-enterprise doctorate).

Research team

People

Research Institute or companies

Department

Role

Dr. E. Tsiporkova

Sirris

Data Innovation

Promotor

Dr. K. De Brabandere

3E NV/SA

iLab

Mentor

Prof. E. Zimanyi

ULB

CoDE

Promotor

M. P. Dagnely

Sirris

Data Innovation

PhD. candidate

Dr. Tom Ruette

Sirris

Data Innovation

Supervisor

Dr. Tom Tourwé

Sirris

Data Innovation

Supervisor

Dissemination

Publication

  • Dagnely, P., Tsiporkova, E., Tourwe, T. & Ruette, T. - Ontology-driven multilevel sequential pattern mining: mining for gold in event logs of PV plants. IEEE Transactions on Industrial Informatics, Special Section on Industrial Sensing Intelligence (submitted)
  • Dagnely, P., Tsiporkova, E., Tourwe, T. Ruette, T. De Brabandere, K. & Assiandi, F. - A semantic model of events for integrating photovoltaic monitoring data. in Industrial Informatics (INDIN), 2015 IEEE 13th International Conference on 24–30 (2015). doi:10.1109/INDIN.2015.7281705
  • Dagnely, P., Ruette, T., Tourwé, T., Tsiporkova, E. & Verhelst, C. - Predicting hourly energy consumption. Can regression modeling improve on an autoregressive baseline? in BNAIC 2015: Proceedings of the 27th Benelux Conference on Artificial Intelligence, Hasselt (2015).
  • Dagnely, P., Ruette, T., Tourwé, T., Tsiporkova, E. & Verhelst, C. - Predicting hourly energy consumption. Can regression modeling improve on an autoregressive baseline? in ECML/PKDD 2015, Proc. of the 3rd International Workshop on Data Analytics for Renewable Energy Integration (DARE’15) (2015).
  • Dagnely, P., Ruette, T., Tourwé, T., Tsiporkova, E. & Verhelst, C. - Predicting hourly energy consumption. Can you beat an autoregressive model? in Proc. of the 24th Annual Machine Learning Conference of Belgium and the Netherlands, Benelearn 2015 (2015).

Presentation

  • Presentation of a poster on 'Predicting hourly energy consumption. Can you beat an autoregressive model?' at the 24th Annual Machine Learning Conference of Belgium and the Netherlands, Benelearn 2015, Delft
  • Live presentation on 'Predicting hourly energy consumption. Can regression modeling improve on an autoregressive baseline?' in ECML/PKDD 2015, 3rd International Workshop on Data Analytics for Renewable Energy Integration, DARE’15 (2015), Porto
  • Live presentation on 'Predicting hourly energy consumption. Can regression modeling improve on an autoregressive baseline?' at the 27th Benelux Conference on Artificial Intelligence, BNAIC 2015, Hasselt
  • Presentation of a poster on 'A semantic model of events for integrating photovoltaic monitoring data. in Industrial Informatics' at the 13th International Conference on Industrial Informatics, INDIN 2015, Cambridge.

Support

With the kind support of Innoviris.