Predicting Archimedes Screw Generator Power Output Using Artificial Neural Networks

Abstract

Previous hydraulic studies of Archimedes screw power generators (ASGs) have been mostly at laboratory scale. The validity of scaling up models based on these studies for application in field-scale ASGs has been a major research gap. This study developed a nondimensional artificial neural networks (ANN) model to predict shaft power of an ASG using extensive multiscale data sets. The model was trained using 583 experimental observations from laboratory-scale and field-scale Archimedes screws over a wide range of volume flow rates, operating speeds, and outlet water levels. The input training data was nondimensionalized to allow for scaling between different size screws. The trained ANN model was used to predict the power output of a different ASG with an average error of 6%. It was found that an ANN can be trained to provide reasonably accurate predictions of ASG power if the training data includes a range of ASG sizes.