Bridging the Temporal Gap: From Historical Monthly Invoices to Granular Hourly Energy Forecasting for Sustainable Operations

Large-scale commercial facilities face significant challenges in sustainability planning due to limited granular energy consumption data. While monthly energy data exists for all facilities in the form of utility invoices, hourly or better resolution is often available for only a subset of locations. This paper presents a machine learning framework that synthesizes low-resolution temporal data to generate high resolution hourly energy forecasts. Our approach combines monthly data from 351 facilities with hourly patterns from 175 instrumented sites using a Bayesian disaggregation model (remaining sites were used as test sites). The system achieves 1-3 month ahead hourly forecasts with a 30% reduction in MAE compared to the utility-bill baseline. This could enable applications in carbon accounting, demand response, and renewable integration planning. As commercial loads continue togrow as a percentage of total grid consumption, forecasting their unique consumption patterns becomes increasingly valuable for grid reliability and efficiency. Grid operators could adopt this methodology to improve the accuracy of their loadforecasts, making the entire grid more efficient and reliable.