An Adaptive Strategy for Wind Speed Forecasting Under Functional Data Horizon: A Way Toward Enhancing Clean Energy

Abstract

An important issue in competitive energy markets is the accurate and efficient wind speed forecasting for wind power production. However, wind speed forecasting models developed for one location usually do not match the other site for various reasons like changes in terrain, different wind speed patterns, and atmospheric factors such as temperature, pressure, humidity, etc. Thus, introducing a flexible model that captures all the features is a challenging task. This paper proposes a functional data analysis (FDA) approach to forecast the site variant wind daily profiles with higher accuracy. Unlike the traditional methods, the FDA is more attractive as it forecasts a complete daily profile, and thus, forecasts can be obtained in the ultra-short period. To this end, the wind speed data is first filtered for extreme values. The filtered series is then divided into deterministic (Component-I) and stochastic (Component-II) components. Component-I is modeled and forecasted based on the generalized additive modeling technique. On the other hand, Component-II is modeled and forecasted using functional models such as functional autoregressive (FAR) and FAR with explanatory variables (FARX). For comparison purposes, forecasts from the traditional univariate autoregressive integrated moving average (ARIMA), seasonal ARIMA (SARIMA), SARIMA with exogenous information (SARIMAX), and neural network autoregressive (NNAR) models are also obtained. For empirical analysis, the wind speed data are obtained from the NASA power project for the site Canada located in Durham, England, and one-day-ahead out-of-sample forecasts are obtained for a complete year. The forecasting performance of different models is assessed through different accuracy measures, namely mean error, root mean squared error, mean absolute error, and mean absolute standard error. The results indicate that the functional models outperform the classical ARIMA, SARIMA, SARIMAX, and a deep learning model, NNAR. Within the functional models, the forecasting ability of the FARX is superior to FAR.

Keywords

• Wind speed
• forecasting
• time series models
• machine learning models
• functional models