The training of pi-sigma artificial neural networks with differential evolution algorithm

In the literature, studies on forecasting problem have started with classical forecasting models and have been successfully continued with artificial neural networks in recent years. In the literature, it is known that many artificial neural networks such as multilayer feed-forward neural networks, multiplicative neuron model neural networks, Pi-Sigma artificial neural networks are very successful in the use of forecasting problem. Pi-Sigma artificial neural networks from these neural networks differ from other neural networks due to the fact that they have both additive and multiplicative functions and are a high order artificial neural network type. In the literature, although the use of heuristic optimization algorithms such as genetic algorithm, particle swarm optimization for the training of Pi-Sigma artificial neural networks used for forecasting purpose, the differential evolution algorithm has not been used yet. Within the scope of this thesis, the training of Pi-Sigma artificial neural networks is carried out for the first time with differential evolution algorithm. In order to evaluate the performance of the proposed method, two different real life time series, which are frequently used in the literature, were analyzed and the performance of the proposed method was compared with some other types of artificial neural networks.