Estimation of long range correlations and FARIMA modelling of wind speed in Maharashtra

Abstract

Modelling and forecasting of solar insolation and wind speed are extremely important in the design and operation of decentralized power generation systems like microgrids. For Indian conditions with varied geographic features, wind speed temporal dynamics are highly characterized by intermittency, due to weather and climatic changes. Modelling of wind speed is done for accurate prediction under different time regimes. Time series models are the most commonly used, for modelling and forecasting due to its simplicity. They easily capture the statistical properties of the data, with medium term forecasting of hours to day ahead with reasonable accuracy. Models such as ARMA and ARIMA have already been used for midterm forecasting for wind speed, and solar insolation data. Analysis of long term temporal and spatial variability of wind speed data is useful for modelling wind related phenomena and quantification of long term wind potential in a location. Detrended Fluctuation Analysis (DFA) is a method to quantify long range correlations in non-stationary time series. This paper describes a Fractional Autoregressive Moving Average (FARIMA) Model applied to a non-stationary wind speed data for a year at a location in Maharashtra. This model combines conventional modelling technique with long term temporal characteristics of the data. Model results provide information related to autocorrelation properties of the given data. The midterm forecasting results are compared with conventional persistence, ARMA and ARIMA Models to highlight the application suitability.