Neural networks are derived to be used as closed-form representations of mean hydrokinetic turbine performance variables. These representations can be used to obtain estimates of turbine performance when the ambient turbulence characteristics, namely in-flow velocity and turbulence intensity, are given. The neural networks were developed using a detailed hydrodynamic code, which simulates performance of a rigidly mounted hydrokinetic turbine where only rotor rotation is allowed (1-DOF). By varying the in-flow velocity (U) of the water current between 0.4m/s and 2.6m/s with a step of 0.2m/s, as well as Turbulence Intensity (TI) between 5% and 20% with a step of 2.5%, a set of variables including the output shaft power, shaft torque, force on a single blade and drag force were obtained for each case. The obtained data sets were used to train appropriately sized, feed-forward (i.e. without recirculation) neural networks. Four neural networks obtained, one for each output variable of the hydrodynamic code. Each neural net constitutes a closed-form, explicit mathematical relationship (equation) generating estimates for the corresponding dependent variable it has been trained to approximate, when presented with specific values for the independent (input) variables current velocity, U, and turbulence intensity, TI. Four output (dependent) variables of interest of the hydrodynamic code are considered: shaft power, shaft torque, force on a single blade and drag. The dependent variables are actually time-averaged steady-state values derived from each hydrodynamic code run. The results of the neural networks are validated using the background theory, as well as the data generated by the hydrodynamic code. Error of less than 1% has been achieved between the neural net output and the hydrodynamics code data values suggesting that the neural networks and the equations are usable in place of the hydrodynamic code for estimating time-averaged loadings and power production.

This content is only available via PDF.
You do not currently have access to this content.