Cluster analysis for classification and forecasting of solar irradiance in Durban, South Africa

Abstract

Clustering of solar irradiance patterns was used in conjunction with cloud cover forecasts from Numer-ical Weather Predictions for day-ahead forecasting of irradiance. Beam irradiance as a function of time during daylight was recorded over a one-year period in Durban, to which k-means clustering was applied to produce four classes of day with diurnal patterns characterised as sunny all day, cloudy all day, sunny morning-cloudy afternoon, and cloudy morning-sunny afternoon. Two forecasting methods were in-vestigated. The first used k-means clustering on pre-dicted daily cloud cover profiles. The second used a rule whereby predicted cloud cover profiles were classified according to whether their average in the morning and afternoon were above or below 50%. In both methods, four classes were found, which had diurnal patterns associated with the irradiance clas-ses that were used to forecast the irradiance class for the day ahead. The two methods had a comparable success rate of about 65%; the cloud cover clustering method was better for sunny and cloudy days; and the 50% rule was better for mixed cloud conditions.