Multifactor Wind Speed Prediction Based On Deep Learning

Accurate wind speed prediction is of great significance in social life.However,it is difficult to achieve high-precision prediction because of the inherent intermittency,volatility,and instability of wind.Existing wind speed prediction algorithms mostly consider the time correlation of wind,but ignore the influence of spatial factors on wind speed,so it is difficult to achieve high-precision prediction.In response to the above problems,taking the 11-year meteorological element record data from 2009 to 2019 provided by Dongying Meteorological Center as an example,this paper constructs an MFSTC model and proposes a multi-factor wind speed prediction method based on deep learning.This method can simultaneously consider multiple spatio-temporal correlations between time,location and meteorological factors,thereby improving the accuracy of wind speed prediction.It is mainly divided into the following three contents.Feature attribute extraction is the primary problem to be solved in multifactor wind speed prediction,which is mainly used to select input samples to reduce the dimensionality of the input variables of the wind speed prediction model.Based on the problem that a single method cannot extract effective feature meteorological elements,this paper combines the PCA and LASSO algorithms to form a new PCA-LASSO algorithm.First,the relevant meteorological factors are transformed into new independent variables through principal component analysis.And then,the coefficients of some meteorological elements are further compressed to 0 through LASSO regression,so as to achieve the purpose of feature selection.This method avoids the adverse effects of redundant factors on the prediction accuracy,and also simplifies the calculation of the model.Aiming at the problem of regional linkage in weather changes,this paper proposes a spatial feature extraction method based on Convolutional Neural Network.First,the feature meteorological elements data is reconstructed in three dimensions(meteorological factors F,space S,and time T)through the MFSTC model,and then the relevant data of all sites are input to the Convolutional Neural Network.After convolution,pooling,and full connection,the spatial correlation between the target site and its neighboring sites is obtained,fundamentally changing the spatial resolution and fineness of wind forecasts.In the numerical weather prediction model,the wind speed changes have obvious time series characteristics.Aiming at the problem that the accuracy of general neural network declines rapidly with the increase of forecast time,this paper uses Long and Short-Term Memory neural networks to extract the time correlation between feature elements,and simulates time delay through sliding time windows.This method can improve the wind speed time series system while retaining the characteristics of the historical series and expanding the amount of data involved in model calculations and reducing forecast errors.This paper conducts experiments on the multifactor wind speed prediction method based on deep learning,and compares and analyzes the accuracy and error obtained of the experiment.Experimental results show that the algorithm proposed in this paper can effectively improve the accuracy of wind speed forecasting and reduce forecast errors.