Abnormal wind speed detection and prediction : methodology and case study

Abstract

Accurate wind speed prediction is crucial for conserving power resources and enhancing power utilization efficiency. However, deviations from typical wind patterns can introduce errors into predictions, potentially leading to imbalances between wind power supply and demand. Consequently, developing a model to forecast abnormal wind speeds is essential. To address this, we leverage the microcanonical multifractal formalism algorithm to detect abnormal wind speeds. In this paper, we integrate ensemble empirical mode decomposition, phase space reconstruction, and long short-term memory (LSTM) networks to predict these anomalies. Initially, wind speed data is meticulously pre-processed to generate datasets for one-hour, one-day, and non-zero wind speeds. Subsequently, LSTM networks are used to forecast abnormal wind speeds. Evaluations of our methodology across different datasets demonstrate its effectiveness, particularly excelling in one-hour forecasts.