Artificial neural networks modelling for power coefficient of Archimedes screw turbine for hydropower applications

Abstract

Hydrokinetic turbines are the most efficient way to generate energy and electricity in hydropower applications. A hydrokinetic turbine's operational characteristics and physical dimensions affect its efficiency. The relationship between the turbine's geometric configuration and output is complicated and nonlinear. Thus, in the current work, a standalone artificial neural network (ANN) with a graphical user interface (GUI) was used to evaluate the performance of an Archimedes screw turbine (AST). This model used the geometrical configuration of the AST as input variables (axle length, blade stride, blade angle, and diameter ratio) and the power coefficient (C-p) as the only output. Among all the neural network topologies, the ANN model with a 4-3-1 architecture generated the lowest average error and root mean square error (RMSE), respectively, of 0.0211 and 0.0008. The predictions of the ANN model were extremely well congruent with available computational fluid dynamics (CFD) and second-order regression model (SORM) data. Additionally, a virtual hydropower system was developed to quantify the effect of AST factors on hydropower production efficiency. The ANN model projections indicate that the diameter ratio is the most sensitive parameter to AST performance, accounting for 84%, followed by blade stride and other factors. The results revealed that the developed model could accurately evaluate the relationship between the geometric configuration of the AST and its hydropower production efficiency.