Firefly algorithm-based neural network for GCPV system output prediction / Nor Syakila Mohd Zainol Abidin

Solar energy is one of the most promising renewable resources that can be used to produce electric energy through photovoltaic process. A significant advantage of Grid Connected Photovoltaic (GCPV) systems is the use of the abundant and free energy from the sun. It is commonly used in urban areas which are readily accessible to the utility grid. However, there are several issues that could possibly slow down the utilization of GCPV system.One of them is the the expected energy output from the GCPV system is unpredictability due to the fluctuating weather conditions throughout the day. Due to this fluctuation, it is difficult for the system owners to identify whether their systems are performing as expected. Due to the importance of predicting the system output, numerous works had been proposed to predict the output from GCPV systems.A few studies had been conducted to predict the output from GCPV systems with Multi-Layer Feedforward Neural Network (MLFNN) being employed as the prediction tool. Nonetheless,a MLFNN is not efficifient enough because it become to be a tedious process and time consuming since it required trial-and-error method for selecting MLFNN training parameters. Therefore, this thesis presents a hybrid Firefly Algorithm (FA)–based MLFNN for predicting the AC power output from a grid-connected photovoltaic (GCPV) system. In the proposed MLFNN, FA was employed as the optimizer and search tools of the MLFNN training parameters. FA was used to optimize the number of neurons in the hidden layer, the learning rate and the momentum rate such that the Root Mean Square Error (RMSE) was minimized. In addition, the MLFNN utilized solar irradiance (SI), ambient temperature (AT) and module temperature (MT) as its inputs and AC power as its output. The performance of the proposed FA-based MLFNN had been compared with the performance of the Classical Evolutionary Programming-based Neural Network (CEP-based MLFNN). The results showed that the proposed FA-based MLFNN had outperformed the CEP-based MLFNN in producing lower RMSE