Recursive Estimates as an Extension to CUSUM-based Energy Monitoring & Targeting

Statistical methods for energy consumption modeling have been static for the past three decades. Energy performance is typically modeled with linear regression and presented for interpretation using cumulative sum of residuals (CUSUM) charts. These mature techniques are general purpose, statistically robust, and simple. They suit energy measurement and verification (M&V) for quantifying a few large performance changes. However, in monitoring and targeting (M&T) applications, CUSUM charts are more challenging to interpret. Energy managers must disambiguate multiple known changes, detect unknown changes and, most importantly, diagnose and act on changes to energy performance. We propose recursive estimates charts as a supplement to CUSUM charts to aid diagnosis in M&T. Recursive estimates charts track changes to linear regression model parameters. Performance changes correlated with model ‘drivers’ (e.g. more gas consumed during cold weather) are reflected as shifts in time-series charts. This provides informative diagnosis guidance to energy managers and can help them be more credible when engaging colleagues in correcting energy waste. We introduce an exponentially-weighted recursive estimate method modified to improve perceptual qualities, and demonstrate its application to an M&T case study.