Asynchronous synthesis of a neural network applied on head load prediction

Abstract

This paper introduces innovative method of an artificial neural network (ANN) optimization (synthesis) by means of Analytic Programming (AP). New asynchronous implementation of Self-Organizing Migration Algorithm (SOMA), which provides effective increase of AP computing potential, is introduced here for time as well as original strategy of communication between SOMA and AP that further contribute towards efficiency in search for optimal ANN solution. The whole ANN synthesis algorithm is applied on the real case of heating plant model identification. The heating plant is located in the town of Most, Czech Republic. The method proves itself to be especially effective when formally identified non-neural parts of the heating plant model need to be made more accurate. Asynchronous distribution plays the key role here as the heating plant behavior data has to be acquired from a very large database and therefore learning of ANN may require a lot of computation time.