Business Objective
Our client is a product resource management company. The client wanted to identify opportunity areas for a leading manufacturer to reduce their energy consumption and carbon footprint across their production facilities.
The client wanted to use historical machine-level production outputs to accurately estimate energy resources required per machine and subsequently take corrective action to reduce energy inefficiencies within their production processes.
Challenges
- Sparse facility output data to identify the right relationships and not over-parameterize or over-fit the model
- The energy consumption at production facilities also depended on several undocumented sources of energy consumption such as HVAC
- Observed temporal relationships between the machines that are generally difficult to capture using sparse data
Solution Methodology
- Worked with the client to understand the available data in the context of the business problem
- Explored the non-linear and temporal relationships of resource consumption with run times and production outputs of various machines
- Identified the right model forms through systematic analyses using training and test datasets, and computed the confidence around our estimates for various machines
- Used bootstrapping to analyze the sensitivity of end-results and assumptions, and refined the algorithms, thus ensuring their robustness
Business Impact
- Our model predicted the overall energy consumption of the production facilities with up to 97% accuracy
- Identified areas where energy efficiency could be gained by modifying the production schedule
