For our industry member Énergir, which is active in the sustainable energy sector, precisely planning the length of connection and main-line work is a critical task, and any errors can drive up the cost of projects.
Margaux Flipo, who at the time was a business analytics student at HEC Montréal, did a four-month internship at Énergir, in partnership with Mitacs, a Canadian non-profit organization that operates this type of research placement, among others, to boost recruiting of highly trained graduates by private-sector players.
Margaux’s assignment, working in collaboration with the Énergir team and under the supervision of Erick Delage, professor in the Department of Decision Sciences at HEC Montréal and member of GERAD, was to develop a predictive tool that would make it easier to estimate the duration of work to build out a gas grid, and in turn improve the accuracy of planning schedules for that work.
She began by making site visits and developing a database using various sources of data gathered over a five-year span that up to then had been used for purely descriptive purposes. The thousands of pieces of data included schedules of work supervisors, the dimensions of areas of land crossed during work, indicators extracted from enterprise resource planning (ERP) software applications, as well as qualitative information.
Next, she conducted trials with the following three methods so as to design an effective forecasting model for estimating work project durations:
- Multiple regression
- Support vector machine (SVM)
- Neural networks
The best models derived from each method were then tested on new samples for the purpose of assessing their actual performance, which proved to be virtually equivalent in each case.
This collaboration with IVADO and Mitacs led to development of a forecasting tool that will help improve our planning processes. Over the longer term, this more accurate estimation of gas grid buildout times should drive greater efficiency across the entire process and eventually lead to lower construction costs.”
Éric Hurtubise, Contract Administration Advisor, Énergir
Lastly, so-called determinist and robust models, each of which had limitations and advantages, were used to perform planning of 19 work sites chosen in the analysis, based on the forecast durations. The robust model turned out to be more representative of real-world conditions, and therefore is likely to enable better management of time-delay risk, which can then be used to optimize upstream allocation of projects among the various teams supervising the work.
Contract Administration Advisor, Énergir