Long-term power forecasting using FRNN and PCA models for calculating output parameters in solar photovoltaic generation

This paper evaluated a 1.4 kW grid-connected photovoltaic system (GCPV) using two neural network models based on experimental data for one year. The novelty of this study is to propose and compare full recurrent neural network (FRNN), and principal component analysis (PCA) models based on entire year experimental data, considering limited research conducted to predict GCPV behaviour using the two methods. The system data was collected for 12 months secondly and hourly data with 50400 samples daily. The GCPV evaluates using specific yield, energy cost, capacity factor, payback period, current, voltage, power, and efficiency. The predicted GCPV current and power using FRNN and PCA were evaluated and compared with measured values to validate results. However, the results indicated that FRNN is better in simulating the experimental results curve compared with PCA. The measured and predicted data are compared and evaluated. It is found that the GCPV is suitable and promising for the study area in terms of technical and economic evaluation with a 3.24–4.82 kWh/kWp-day yield, 21.7% capacity factor, 0.045 USD/kWh cost of energy, and 11.17 years payback period. •Predict grid-connected PV system output using principal component analysis and recurrent neural approaches.•Use one-year measured data to validating the proposed models.•Evaluation and comparison of system performance using ANN models and experimental results. Grid connected PV; recurrent neural; principal component analysis; Desert type PV; ANN.