Abstract
Site suitability analysis with geographical information system multi-criteria decision-making (GIS-MCDM) techniques forms a significant process in wind and solar energy exploration at the utility-scale level. The process unveils viable sites for exploration, however, few is known about the variability investigation of these sites before physical site development. Besides, soft computing techniques like Adaptive Neurofuzzy Inference System (ANFIS) at standalone and hybrid with population-based optimization models like Particle Swarm Optimization (PSO), and Genetic Algorithm (GA), have been effective in understanding the variability and intermittency in wind and solar resources. Hence, their integration with GIS-based site suitability analysis for variability investigation in viable sites for utility-scale wind and solar resources offers a high potential for strategic and operational resource planning. Pilot studies that investigate the effectiveness of the developed standalone ANFIS, GA-ANFIS, and PSO-ANFIS models were carried out and all these present PSO-ANFIS as highly effective, however at a higher computational time(CT). Furthermore, site-specific investigations on wind and solar resources predictions using standalone ANFIS and PSO-ANFIS models were performed. First, the significance of data clustering (Fuzzy-c-means (FCM), Subtractive Clustering (SC), and Grid Partitioning (GP)) on the two models using wind power time-series data was investigated. The SCbased PSO-ANFIS model performed best among the three models with root mean square error (RMSE)= 0.127, mean absolute deviation (MAD) = 0.078, mean absolute percentage error (MAPE)= 28.11, relative mean bias error (rMBE) of 0.190 and Variance Accounted For (VAF) of 94.311, and CT= 47.21 secs. Second, the effectiveness of GA and PSO-based ANFIS models on power forecast for three wind turbine generators(WTG) was performed and the PSO-ANFIS model performed better on the first WTG with RMSE= 0.180, MAD= 0.091, R2= 0.914, and CT= 702.3 secs. In the same study, further investigations on the feasibility of embedded generation for powering a nearby agricultural farm was carried out. Similarly, wavelet-based ANFIS models were investigated on solar radiation forecast and it was established that standalone models performed better than wavelet hybrids, though many factors can be responsible for this. With the effectiveness of the soft computing model assured, they were used for resource variability investigation in candidate viable sites obtained from GIS-based site suitability analysis for wind and solar energy in the Eastern and Western Cape Provinces respectively. Twenty years historical satellite data for wind and global horizontal irradiance (GHI) obtained from the National Aeronautics and Space...
Ph.D.