Operative estimation of global horizontal irradiance using transfer functions through network types of artificial neural network in some selected sites in North-East Ethiopia: assessment and comparison

The estimation of global horizontal irradiance (GHI) for a specific location is crucial to the design, modeling, and effective operation of photovoltaic (PV) and solar thermal energy conversion systems. This study aims to predict the monthly average daily and monthly average GHI for three locations in Ethiopia (Kemissie, Haik, and Woldiya) using artificial neural networks (ANNs) with various activation functions. Days of the year, elevation, and average values (maximum and minimum temperatures, surface pressure, wind speed, relative humidity, wind direction, and precipitation) were fed into the ANN during training. For the prediction of monthly average daily GHI, the CFBP-tansig, FFBP-tansig, and FFBP-logsig networks proved to be the most accurate. Likewise, for monthly average GHI prediction, the FFBP-tansig, CFBP-tansig, and LR-tansig networks exhibited the best performance. The daily and monthly GHI predictions achieved RMSE values ranging from 0.00058 to 0.04042, with correlation coefficients (r) between 0.95417 and nearly 1.0. In comparison, traditional methods showed RMSE values between 0.1 and 0.2, indicating that the ANN models provided up to 80 % better accuracy. The EBP-purelin, CFBP-purelin, and LR-purelin models were less effective, with RMSE values between 0.13815 and 0.17307, and r from 0.91153 to 0.96719. Among the studied locations, Kemissie exhibited the highest GHI potential, emphasizing the region's significant capability for solar energy generation in Ethiopia.