Wind Power Pattern Prediction in time series measuremnt data for wind energy prediction modelling using LSTM-GA networks

Wind Power forecasts in the short and medium term time horizons are essential for overcoming variable energy productions caused by the changes in weather conditions and fluctuations in wind speed. Wind speed follows a time series pattern in time that can be exploited to forecast for power production. This paper uses a long short term (LSTM) neural network model to predict and validate wind power using time series measurements of an online available wind power measurements. Short term pattern prediction (for accuracy) and mid- term (for stability) are studied for wind power prediction. To increase the reliability of the model a genetic algorithm is used to optimize the window of time capture to alleviate problems of overfitting or an appropriate window size that does not lead to any loss of information. A root mean square error of 0.0957993 and 0.0929905 is obtained for the short term and medium predictions respectively.