Inverse neural network to optimal performance of the hydraulic turbine runner blades

Abstract

Inverse Artificial neural network (iANN) is applied to optimize the parameters for hydraulic turbine runner blades. This proposed method (iANN) inverts the artificial neural network (ANN) and uses an optimization method to find the optimum parameter value (or unknown parameter) for given required conditions. In order to do so, first an ANN, which consists of a feedforward network with one hidden layer that simulates the efficiency of the turbine runner blade, was developed. This model takes into account parameters of the turbine as well-known input values. For the network, the LevenbergMarquardt learning algorithm, the hyperbolic tangent sigmoid transfer-function, the linear transferfunction and three neurons in the hidden layer were considered to predict a satisfactory efficiency (R>0.99). As for the validation of the data set, simulations and target tests were in good agreement (R>0.99). Second, results of the iANN showed also a good agreement with target and simulated data (error < 1.5%), in this case, the optimum mass flow was calculated and validated experimentally to a given required efficiency (95% and 99%). Then iANN could be applied to determine the optimal parameters on the hydraulic turbine runner blades with an elapsed record time: minor to 0.1 seconds. Another result of this research is the interesting advantage provided by this methodology as it may be used to find out any input parameters.