New Zealand Wind Speed and Power Potential Forecast: A case of Nelson

Authors:

Abstract
This paper introduces a novel framework for wind speed forecasting in Nelson, New Zealand, leveraging meteorological insights alongside advanced deep learning methods. A transformer-based model with multi-head attention, custom cyclical temporal encodings, domain knowledge features, and specialized loss functions is employed to predict 24-hour wind speeds at 10-minute intervals. The resulting forecasts attain a Mean Absolute Error (MAE) of 0.35 m/s and a Mean Squared Error (MSE) of 0.24, aligning competitively with leading forecasting techniques. To further enhance applicability, the model integrates air density calculations for precise wind power estimation using an actual wind turbine power curve, yielding an average power MAE of 13.27 KW. The proposed approach shows considerable promise for renewable energy planning toward 100% renewable electricity in New Zealand.

This publication has 14 references indexed in Scilit: