Machine Learning Driven Performance Benchmarking for Energy Efficiency

Becoming ‘Carbon Neutral’ in the coming decades is increasingly a focus for many large organizations as the world recognizes the need for sustainable energy. Two prominent ways this can be achieved are reducing energy consumption by efficient energy utilization and shifting to renewable sources of energy. To address them, we have come up with a framework which can be used for large US-based big box retailers. The proposed solution aims to benchmark sites (and subsystems) laterally and longitudinally to identify underperforming sites, conduct audits and improve energy efficiency. To achieve these objectives, our solution learns energy consumption pattern for every retail & nonretail sites and different subsystems (like - HVAC, Lighting and Refrigeration).

The outcome of benchmarking can also help us prioritize sites to focus on new initiatives like replacing conventional lights with LED (Light Emitting Diode) lights, finding energy efficient sites for solar power installations. An additional outcome of the solution is the ability to provide futuristic (after 25-30 years) view of energy load profile for each site, that caters for an important insight while planning the infrastructure as retail corporations move towards becoming carbon neutral. The estimated baselines (model predicted load profile) can also cater to other use cases, such as investigating large energy consumption deltas leading to probable larger issues such as poor asset health, automated dimming in lighting not working, wrong set points for assets, operational issues like outdoor lighting always on or using z-statistic to compare the actual and estimated energy consumption and generate potential fault alarms etc.