Investigating the trade-off between response time and complexity in the Levenberg–Marquardt ANN-MPPT algorithm used in wind energy conversion systems

Abstract

The integration of artificial intelligence (AI) models in renewable energy resources management, particularly in the utilization of maximum power point tracking (MPPT) optimizers, has gained significant attention. This study focuses on investigating the tradeoff between accuracy, response time, and system complexity by varying the number of neurons in artificial neural network (ANN) models for MPPT in wind energy conversion systems (WECSs). Traditionally, MPPT algorithms in WECSs are implemented using direct or indirect methods. However, these methods lack an accumulative learning curve and rely on instantaneous inputs. In contrast, ANN models trained on pre-existing datasets offer the potential for improved maximum point capturing processes. Nevertheless, the incorporation of ANN models may introduce additional complexity to the system. Two ANN models, direct and indirect, are examined in comparison to a reference model using the perturb and observe conventional MPPT algorithm. The results show that the ANN direct model exhibits better time response in the face of high variations in wind speed profiles. On the other hand, the ANN indirect model demonstrates a 4% increase in accuracy with minimal ripples.