Neural Network Model for Elevator Dynamics

Neural Network Model for Elevator Dynamics

Abstract





Ride comfort has an increasing role in the elevator business. Ride comfort is a result of both the product and also professional skill in elevator installation, servicing and maintenance. Computer aided simulation can be used in obtaining valuable information of the ride comfort for use in development. A good model can be used in off-line control design and implementation of new advanced control schemes.
In this paper a Nonlinear recurrent Output-Error (NOE) neural network model for elevator dynamics is developed. The model combines nonlinear and linear network into a Grey-box model instead of common Black-box neural network model. Physical knowledge is embedded into network construction via creating a sparse network.
Both theoretical knowledge and practical engineering aspects must be considered in industrial applications. A model must be found which combines both robustness and accuracy in desired extent as well as is computationally light and cheap to be applied in mass product systems. To succeed in filling these requirements a great deal of background work for identification needs to be done.
This paper is based on work done in a Dynamic System Modelling (DYSMO) subproject of Finnish Technology Development Centre's Adaptive and Intelligent Systems Applications technology programme. Neural computing and fuzzy logic are being exploited in the programme.