Predicting Global Solar Radiation using Gamma Test and Local Linear Regression Data Models in Bauchi, Nigeria

"We report the implementation of a new perspective in non-linear modelling resulting in the development of data models for prediction of global solar radiation () in Bauchi (10’.19’ N, 9.51’ E) Nigeria. Several studies to find alternative means of generating good estimates of data have been carried out and have continued for locations in Nigeria including Bauchi. Questions relating to the amount of data, determination of possible input combinations and noise variances before modelling has not been answered. Using Gamma Test(GT), weather variables namely, rainfall (), cloud cover, average temperature(), average relative humidity (), wind speed (U), extraterrestrial radiation () and cleanness index () as input variables and as the output in the numerical data set named bauchi-solar.csv for a period of five years (2003 ? 2007) with 853 unique data points() were examined. An optimal near neighbour number () of 32 was used in arriving at a gamma statistic, of 0.0013. The M-test analysis shows that the noise variance becomes relatively stable at between 700 and 750 data points being data required for the modelling process. 127 Local Linear Regression (LLR) models were identified using the Full Embedding Search heuristics with coefficient of determination () values for both the training set (1-750) and the validation set (751-853) ranging between -18.85 to 0.9923 which guided the selection of models. The average MSE for both training and validation data sets are between 0.0254 and 20.77 for the location. An average value of about 0.99 for the gradient showed that the functions were generally moderate in complexity. "