Prediction of soil temperature by using artificial neural networks algorithms

In this paper we make use of neural networks algorithms to predict the soil temperature from the known previous data. We first consider a single layer neural network having m McCulloch- Pitts Type neurons and use the generalized Widrow-Hoff algorithm to train the network. We give conditions on the learning rate and the transfer functions which will guarantee the convergence of the generalized Widrow-Hoff algorithm. To prove the convergence we make use of Fixed- point theorem. Our convergence theorem generalizes an earlier convergence theorem proved by Hui and Zak. We also consider multi-layer neural networks for the prediction where we use back-propagation algorithm with momentum for training the networks. The data used for training is taken from the observatory of the department of Agriculture Meteorology, B. A. College of Agriculture, Gujarat Agricultural University, Anand, Gujarat, India for the year 1999.