Steam versus Hot-Water Campus Heating: A Data-Driven Life-Cycle, Reliability and Decarbonisation Analytics Framework for the Phased Conversion of University District Energy Systems

Abstract

Many universities still heat their campuses with central steam plants and networks built in the first half of the twentieth century, and the case for converting them to low-temperature hot water is usually made with single-point figures of weak provenance. This article designs a data-driven decision-analytics framework for the steam-versus-hot-water decision and evaluates it on a transparent parametric model of a synthetic campus: 60 buildings, 560,000 m², a 9.9 km network with segment ages of 13–85 years, four climate archetypes and six strategies — steam with reactive or programme maintenance, conversion in three or nine years, each with or without a heat-pump plant — under 3,000 Monte Carlo draws of 32 uncertain inputs over 30 years. The framework links a calibrated distribution-loss model, a network model with trap, condensate, flash and blowdown balances, an age-dependent failure hazard, probabilistic life-cycle costing with phasing and parallel operation, Sobol sensitivity analysis, a metamodel and measurement-and-verification feedback. Documented campus conversions were verified against primary sources and used to bound the model; several figures in circulation could not be verified. On the model, steam distribution loses 20–35% of plant output and delivers 49–62% of fuel as building heat against 87–88% for hot water, so conversion cuts gas use by 29–37% and 30-year emissions by 26–33% (46–50% with heat pumps). Its cost advantage is conditional: fast conversion is least-cost in 33% of draws in the warm climate and 78% in the cold one, rising to 90–100% with a carbon price of 190 per tonne or a small internal-retrofit scope and falling to 7–46% if the mains are 20 years younger. Asset age, discount rate, retrofit scope and gas price explain most of the variance; a nine-year programme costs 15–18 per m² more than a three-year one because the steam plant, its staff and both networks run in parallel. Trap monitoring beats periodic surveys at all gas prices. All inputs are stylised with stated ranges; the model and results file are provided.