Identification of the optimal control characteristics of a small hydropower plant using artificial neural networks and the support vector machines method

• Published in: Journal of hydraulic research Vol. 57; no. 5; pp. 715 - 723
• Main Author: Borkowski, Dariusz
• Format: Journal Article
• Language: English
• Published: Madrid Taylor & Francis 03-09-2019; Taylor & Francis Ltd
• Subjects: Approximation; Artificial neural networks; Control structures; field studies; flow control; Flow rates; Flow velocity; Fluid flow; hydraulics of renewable energy systems; Hydroelectric plants; Hydroelectric power; Hydroelectric power plants; Identification; Mathematical analysis; Multilayers; Neural networks; Optimal control; Procedures; Radial basis function; Support vector machines; Training; Turbine engines; Turbines; velocity measurements
• ISSN: 0022-1686; 1814-2079
• DOI: 10.1080/00221686.2018.1522378

This study investigates a small hydropower plant at varying turbine speeds. The identification of the optimal turbine rotational speed as a function of a flow rate is usually based on the efficiency function of two variables. The existing procedures require a large number of operating points, which prolong the necessary field measurements. In this paper, a new identification procedure based either on the use of a traditional neural network (multi-layer perceptron, radial basis function), or on a support vector machines method using dedicated assessment factors, is proposed. The procedure was tested and verified in an experimental 150 kW small hydropower plant that contained two propeller turbines. The tests highlighted the advantages of the support vector machines network over traditional neural networks owing to the precise approximation under a limited amount of training data. It is shown that even 15 measurement points provide relatively accurate results.