Soil Temperature Prediction Using Time-Delay Neural Networks

Abstract

AbstractNeural networks are widely used for time series prediction in the recent years. In particular, dynamic neural networks with embedded time delays are the most appropriate models for the simulation of nonlinear processes since they make use the effect of past input values. The purpose of this study is to predict soil temperature in various depths, by using dynamic neural networks. The dynamic networks used are recurrent neural networks with feedback loop that includes time-delay elements. The data used for the neural network’s training, validation and testing were hourly values obtained from the weather station at the Agricultural University of Athens, for the period 2002–2005. Error statistics of the results showed a good fitting of the models.KeywordsSoil TemperatureMean Square ErrorRecurrent Neural NetworkRecurrent NetworkNeural Network ToolboxThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.