Skip to main content
Article
Effect Of Different Mappings And Normalization Of Neural Network Models
Ninth National Power Systems conference, Indian institute of Technology, Kanpur (1996)
  • D. K. Chaturvedi, Dayalbagh Educational Institute
  • P. S. Satsangi, Dayalbagh Educational Institute
  • P. K. Kalra
Abstract
Feedforward Neural Network Using supervised learning has been well accepted for modelling of processes, plants and predictors. This model needs many pairs of input-output (X-Y) as training data. The relation between input and output, the size of neural network, type of neuron, and connectivity of neurons among various layers generally contribute to training time of the neural network. Results have been reported to reduce the training time by modifying learning algorithms for the neural networks. This paper presents cases where it has been observed that training time is a function of normalization, thresholding function and type of relationship of input-output pairs. The relationship can be generated in terms of variables or change of variables. Hence, the study reported here, consider X-Y and type of relationships. In addition, influence of noise in the input-output data an accuracy of learning and training time has een studied. The neural network model has been developed to study above mentioned issues for DC machines to predict armature current and speed, and for short term load forecasting problems.
Keywords
  • ANN,
  • Mapping,
  • Scaling,
  • Threshold function,
Publication Date
December, 1996
Citation Information
D. K. Chaturvedi, P. S. Satsangi and P. K. Kalra. "Effect Of Different Mappings And Normalization Of Neural Network Models" Ninth National Power Systems conference, Indian institute of Technology, Kanpur Vol. 1 (1996)
Available at: http://works.bepress.com/dk_chaturvedi/27/