Day-ahead prediction using time series partitioning with Auto-Regressive model

TL;DR

This paper introduces a novel approach combining time series partitioning with Auto-Regressive models to improve day-ahead wind speed prediction accuracy, addressing the challenge of accurate wind forecast for efficient wind farm operation.

Contribution

It proposes a new method of data partitioning and forecasting that enhances wind speed prediction accuracy over traditional approaches.

Findings

Significant improvement in wind speed forecast accuracy.

Effective use of time series partitioning with AR models.

Enhanced prediction performance for wind farm management.

Abstract

Wind speed forecasting has received a lot of attention in the recent past from researchers due to its enormous benefits in the generation of wind power and distribution. The biggest challenge still remains to be accurate prediction of wind speeds for efficient operation of a wind farm. Wind speed forecasts can be greatly improved by understanding its underlying dynamics. In this paper, we propose a method of time series partitioning where the original 10 minutes wind speed data is converted into a two-dimensional array of order (N x 144) where N denotes the number of days with 144 the daily 10-min observations. Upon successful time series partitioning, a point forecast is computed for each of the 144 datasets extracted from the 10 minutes wind speed observations using an Auto-Regressive (AR) process which is then combined together to give the (N+1) st day forecast. The results of the computations show significant improvement in the prediction accuracy when AR model is coupled with time series partitioning.