Techniques for Training Sets Selection in the Representation of a Thermosiphon System Via ANN

The research of alternative forms of energy production became more important nowadays in a context where the natural resources are scarce. In this sense, thermosiphon systems have been developed as an alternative way of energy economy for the water heating process using a renewable energy source: the sun. A thermosiphon system is greatly influenced by several parameters: the ambient temperature, the input water temperature, the solar irradiance, the flow rate, the inclination of the solar collector, the height of the water storage tank and mainly by the manufacturing process. Nowadays, there are interests in the development of analytical models that consider parameters of installation such as: height of the water storage tank and inclination of the solar collector. These analytical models can be complex and nonlinear. In the last decades, ANN's (i.e. artificial neural networks) have been used to represent many kinds of industrial processes, dealing with the complexity and non-linearity of them. Moreover, ANN's are capable of dealing with aspects of manufacturing not considered by the analytical models but that are important in determining the efficiency of the real thermosiphon system. A trained ANN can eliminate the necessity of new laboratory experiments for real and new conditions of installation. A better modeling of the process by means of ANN depends on a representative training set. In order to define the training set better, statistical ways and clustering techniques have been proposed and compared. Results of both techniques have been discussed in this work.