Tethered Balloon Wind Probability Prediction Based on LSTM—QRNN

Probability distribution of the wind around the tethered balloon is one of the important reference factors for tethered balloon systems deployment and daily operation. This paper presents a novel method to forecast the densities of the wind around the tethered balloon, considering the complex physical significance and non-stationary characteristics in wind uncertainties. A Long Short-Term Memory (LSTM) network is utilized to model the long short-term dependencies of the sensor data in order to achieve a deterministic prediction of the perimeter of the wind field where the system is located. To quantitatively characterize the uncertainty of the predictions, a neural network guided by a loss function combining Huber loss and pinball loss(QRNN) is utilized to fit the quantile distribution of the predictions The resulting distributions are cross-validated to determine the optimal bandwidth for kernel density estimation so as to obtain a nonparametric estimate of the wind field probability density function. By comparing with the traditional method, the experimental results show that the proposed method has significant advantages in the task of wind field probability prediction, and can provide more reliable decision support for wind energy resource assessment and risk management in related fields.