Short-term average wind speed and turbulent standard deviation forecasts based on one-dimensional convolutional neural network and the integrate method for probabilistic framework

摘要

Accurate wind speed forecast can provide important information for power system dispatching. Many studies focus on this topic over the last decades, but it's consistently a tough issue due to the intense uncertainty of wind. In physical perspective, wind speed consists of average component and turbulent component. Therefore, oriented to actual demand and improve the forecast precision, this paper develops a novel data-driven method to realize short-term combination forecasts of average wind speed and wind turbulent standard deviation. According to atmospheric boundary layer theory and correlation analysis of the two prediction targets, their time-delay items are combined as the model input features. One-dimensional convolutional neural network is innovatively applied for this work to excavate the timing coupled information in data. Well-performed models can be established after adequate training and validation. Inspired by Pauta criterion, adaptive parameter named as "turbulent standard deviation multiplicator" is defined, which is the specific value of predicted average wind speed error and wind turbulent standard deviation. It is decided as the medium to extend the study to probabilistic framework. Based on its quantile analysis, the statistical significance of the parameter is verified and the prediction results can be integrated to achieve 4 h ahead probabilistic wind speed forecasts. Actual data from China wind farm is utilized to execute the case experiments. Superior performances indicate the feasibility and effectiveness of proposed method and the uncertainties of wind are better learned.