Weakly Supervised Learning of the Motion Resistance of a Locomotive Powered by Liquefied Natural Gas
Abstract
A model of the running resistance of a locomotive powered by liquefied natural gas is proposed. The model uses operating data and does not require specific instrumentation. The input data consists of a succession of instantaneous speed and electrical power measurements of a diesel-electric locomotive. The slope at each point along the route is unknown and the speed is measured with a digital sensor that quantifies the signal, so acceleration estimates are also unreliable. From these data, a weakly supervised learning problem is defined that makes use of a fuzzy rule-based system to indirectly predict the effective slope, and is able to estimate the power demand of the locomotive with a margin of error close to 5%.
Type
Publication
In International Workshop on Soft Computing Models in Industrial and Environmental Applications
