This paper presents different prediction models for a grid-connected photovoltaic (GCPV) system based on long-term data sets. A 1.4 kW GCPV installed in Sohar, Oman, where measurements of the electrical and environmental parameters are taken every second for 5 years (from May 2014 to April 2019) to reduce uncertainty and improve the proposed model's accuracy. The highest power and energy measured from the GCPV are 1180 W and 245.8 kWh, respectively. Mathematical regression and cascade-feed forward propagation (CFNN) models for PV current, voltage, and power were proposed in terms of environmental impact. A NeuroSolutions package was used to propose GCPV models for electrical quantities. An evaluation criterion is defined, in this study, to assess the system's performance. The proposed models show excellent agreement with the measured experimental data. However, CFNN shows a superior accuracy compared with empirical and regression proposed models. The two models have proposed a dispute, but the model relating the solar irradiation and ambient temperature to GCPV current is more accurate. To evaluate and validate the proposed CFNN models, mean square error, MAE, root mean square error, and R-2 metrics have been used as a criterion and compared with different artificial neural networks models in the literature. The proposed CFNN is found to give the most accurate results in terms of MSE = 0.0007, MAE = 0.4310, and RMSE = 0.0290 and highest accuracy R-2 = 0.9999, which shows the effectiveness and feasibility of the proposed models.