A neural network-based wind forecasting model for wind power management in Northeastern Thailand

This paper develops a wind forecasting model to be used for wind power management in northeastern Thailand. The neural network method is employed. Neural networks are trained and evaluated using observations from 17 wind stations in the region. Two forecast times, i.e., 3 and 6 hours in advance and 2 altitudes, i.e., 65 and 90 m above ground are considered. Inputs to neural networks include observed wind speeds at 24 consecutive hours prior to the forecast time at the forecast wind station. The training data for neural networks include more than 174,000 samples in years 2011 and 2012. The forecast accuracies are evaluated using more than 83,000 samples in year 2013. Ten different neural networks are trained for each task and the best neural network is chosen. Forecasts agree well with observations. The 3-hour forecasts are more accurate than the 6-hour forecasts. Forecasts are positively biased for wind speeds below 4 m/s and are negatively biased for wind speeds above 4 m/s. Forecasts for both forecast times at both altitudes have good utility for wind speeds above 2 m/s. The neural network-based wind forecasting model presented in this paper can provide useful wind speed forecasts in northeastern Thailand and can be adapted for other regions.