Modelling and forecasting for solar irradiance from solarimetric station

Abstract

This paper proposes two approaches for modeling solar irradiance time series. The first is an exploratory analysis and second is a periodic forecast model for solar irradiance data from solarimetric station in a country city of Bahia, Brazil. Two normality hypothesis tests were applied, Anderson-Darling and Shapiro-Wilk. The shape and symmetry of the data were also analyzed using boxplot and histograms to investigate the extreme points, as well as great asymmetry in the distribution of data at different times. Through adherence tests, the normal distribution of solar irradiance in most hours within the annual periodic is rejected. A study by season was carried out, which showed a different behavior in relation to the symmetry of the data, with autumn being the most uniform and spring being the most stochastic. Spring and summer presented better conditions for installation of photovoltaic plates, due to the high solar irradiance rates, and autumn and winter presented satisfactory solar irradiance to maintain this form of generation throughout the year.The second approach is a forecast to solar irradiance in photovoltaic generation systems. The importance of this forecast type is to favor the routine of planning, operation and maintenance of these types of systems, in addition to serving as a basis for feasibility studies and expansion of solar generation. Given a stochastic resource and the periodic behavior of solar irradiance data, a periodic autoregressive model was considered; the statistics were used by the maximum likelihood method, based on hourly measurements of irradiance over a period of one year.