Running Resistance Identification for a MAGLEV Vehicle using Successive Learning.

Abstract

A magnetically levitated (MAGLEV) vehicle propelled by a linear synchronous motor (LSM) is one of hopeful candidates to be used for high speed transportation in the next generation. The vehicle will be automatically controlled from start to stop by a control center. In order to increase the ride quality of the vehicle, feed-forward control of the running resistance is added to an ordinary feed-back speed control to compensate disturbances due to fluctuation in a running resistance.The running resistance is also required to predict a current reference to the next substation which covers a section into which the train travels. The running resistance of the MAGLEV vehicle is modeled as a function of speed and is used for feed-forward control and calculation of the estimated current in the substations. The modeling requires identification of the coefficients of the function to specify the running resistance.The formulation of these coefficients has been studied in theory in other literature. No effort, however, has been reported on their quantitative estimation with the use of experimentally gathered running data. This paper proposes a method for identifying these coefficients by successive learning. In this process, these coefficients are modified by means of the method of steepest decent to minimize the error between measurement and estimation of the function. Systematic learning can therefore be realized.Its usability is confirmed with the data gathered at the MIYAZAKI test track.