Advanced Time Series Forecasting Models for Wind Power Prediction and Grid Stability

Abstract

Reliable wind power forecasts are necessary for ensuring the electric grid's reliability and for integrating the intermittent renewable power sources into it. Although several models for forecasting time series are already available, a methodical and empirical analysis of these models' comparison under the same operating conditions is not available. In this paper, compare Autoregressive (AR), Autoregressive Integrated Moving Average (ARIMA), ARIMA with Exogenous variables (ARIMAX), and Seasonal ARIMA with Exogenous variables (SARIMAX), four models by means of a rigorous multi-site evaluation approach with 1795 daily averaged data points for each location from four sites dispersed across a geographic area. For performance evaluation, employs Normalized Mean Absolute Error (NMAE), Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Mean Squared Error (MSE). The results show that incorporating a meteorological exogenous variable provides a boost in prediction accuracy to the models. By adding meteorological information, prediction accuracy gains in stable meteorological regimes can reach 50%-70% in the reduction of the forecast error. From the relative error comparison, the SARIMAX model achieved better prediction accuracy and reliability, and it also captured the weekly cycle and fluctuations in high-value states, compared to baseline models, which achieved NMAE 0.264 at Location 4. The presented clear hierarchy in prediction accuracy may allow grid operators to make better decisions to commit units and dispatch economically based on observed evidence. In summary, linear time series models work quite well under the scenario with stable meteorological conditions, whereas site-specific volatilities demand adaptive, high-accuracy forecast models.