Poorly cooling district-heating installations raise return temperatures and pumped flow, increasing heat losses and pumping energy and hindering low-temperature operation. Faults often go unreported for weeks, so their onset is unknown and energy is wasted in the meantime. Utilities flag poorly cooling installations by a low supply-return temperature difference, but not when the problem began. Changepoint detectors such as the Bayesian Estimator of Abrupt Change, Seasonality, and Trend (BEAST) can date a change but report only its probability, so a large, fault-like change and a marginal one can look alike. We therefore develop a decision score that weighs each BEAST change in temperature difference and flow by its probability, size, deviation from recent operation and physical direction. Joint falls in temperature difference and rises in flow, the signature of insufficient cooling, rank first. Across 1,522 cases, 648 with a reported fault, 24 of the 25 highest-ranked cases were reported faults, and Average Precision was 0.772 against a prevalence of 0.426, likely conservative, as unreported cases may contain faults. Without using the report date, the strongest flow rise in reported-fault cases came a median of 11 days before the report (128 days in comparison cases). Combined with temperature-difference screening, the score raised Average Precision in season-matched cases from 0.486 to 0.646. After the detected change, flagged reported-fault installations returned water 16 K warmer and about half their circulated water delivered no heat. The score gives utilities a short list for inspection and a candidate onset date for each change.
翻译:暂无翻译