Application of an operational meso-scale modelling system for industrial plant energy operations

In our continuing work on the implementation and applications of an operational mesoscale modelling system dubbed "Deep Thunder", we examine its application to energy management of industrial plant operations. The Deep Thunder system has provided 24-hour forecasts for several metropolitan regions in the United States for a number of years. Model forecasts, are typically updated twice daily with triply, nested grids down to one to three km resolution. Explicit, bulk cloud microphysics are included in the model predictions to enable forecasts of potentially severe weather. All of the processing, modelling and visualization are completed in one hour or less on relatively modest hardware to enable sufficiently timely dissemination of forecast products for potential weather-sensitive applications.

Some of the recent extensions to Deep Thunder have included support for weather-sensitive operations at several major IBM facilities in the northeastern United States. These facilities range from large office complexes to research facilities and semi-conductor manufacturing plants. As part of the focus and evolution of information technology-driven environmental analytics, we are investigating the application of high resolution numerical weather prediction models in optimizing energy peak load shedding and free cooling strategies. The location of the IBM Burlington, Vermont facility presents challenges for weather model prediction as a result of the complex orography and local climate, but also offers significant opportunity for free cooling and peak load shedding. The weather model provides highly granular predictive information for use in multivariate site electrical load shedding and HVAC free cooling algorithms. The operational NWP system includes local-scale verification of forecasted meteorological variables by comparison to surface observations. Evolution of the verification system will incorporate available observations from local mesonets to yield more reliable results. This offers potential improvement via larger sample size, and more granular spatial and temporal coverage especially over complex terrain. In addition to site-specific forecast customizations and integration, improvements could include extended forecast periods beyond 24 hours to address operational dependencies as well as incorporation of direct assimilation of the aforementioned observations and adjustment in the physical parameterizations to improve forecast quality. As part of the work, we are attempting to quantify the potential for improving energy and cost savings for a large industrial site. We will discuss our overall approach to these issues and some of the results to date.