Efficient Maximum Power Point Tracking Techniques for a Grid-connected Photovoltaic System using Artificial Intelligence

Abstract

The maximum power point tracking (MPPT) technique is considered a crucial part in photovoltaic (PV) system design for maximising the output power of a PV array and improving the stability and reliability of the PV system. This research focuses on developing common MPPT techniques, including: perturb and observe (P&O), fuzzy logic control (FLC), an adaptive neural-fuzzy inference system (ANFIS) and an artificial neural network (ANN) for a grid-connected PV system, with the best of them being identified. Whilst several techniques have been designed, the P&O algorithm is widely used for MPPT due to its low cost and simple implementation. However, the main drawbacks of this method are a slow tracking speed, high oscillation and a drift problem associated with changing irradiance rapidly. Hence, a modified P&O-MPPT based on the Pythagorean theorem and a constant voltage algorithm is proposed to address those issues by developing variable step size and early step decision for the conventional P&O algorithm, respectively. Unlikely, these modifications do not avoid the drift problem nor eliminate the oscillation completely. The FLC is a commonly deployed technique that achieves vastly improved performance for the MPPT technique in terms of response speed and low fluctuation. However, the key issues of the conventional FLC-MPPT are the drift problem and complex implementation, when compared with the P&O-MPPT. Hence, the MPPT technique based on the FLC and P&O algorithm is proposed to address these challenges. This technique incorporates the advantages of the P&O-MPPT to account for slow and fast changes in solar irradiance as well as reduced processing time for FLC-MPPT to address complex engineering problems when the number of rules of membership functions are fewer. As a result, the proposed technique achieves average tracking efficiencies of around 99.6% under the EN50530 standard test. The ANFIS technique and the ANN technique are used to predict the maximum power point of a PV array, using experimental training data, instead of the rules of membership functions. To improve the accuracy of those techniques, a curve fitting technique and the Particle Swarm Optimisation algorithm are utilised, respectively. These optimisations are classified into two strategies: adjusting the tuning of the ANFIS model as well as determining the right topology and the initial weights of the ANN model. As a result, the training errors of those models are minimised. Hence, the ANFIS technique achieves average tracking efficiencies of greater than 99.3% under a semi-cloudy day test, while the ANN technique delivers average tracking efficiencies of more than 99.67% and 99.30% on sunny and cloudy day tests, respectively.