Application And Research Of Time Series Forecasting In Wind Farm

As a new clean energy source,wind energy has attracted much attention due to its easy access and pollution-free characteristics.The intermittency and uncontrollability of wind bring difficulties to wind power grid connection and even endanger the stability of power system.Therefore,accurate prediction of wind speed and wind power in wind farms plays a crucial role in maintaining the safety of the power system.Due to the complex weather conditions and human factors,achieving high accuracy wind speed and wind power prediction is a challenging task.In order to improve the accuracy of prediction,this paper investigates wind speed and wind power prediction in wind farms.The main work of this paper is as follows:(1)A wind speed prediction model based on a decomposition combination is proposed.Due to the nonlinear and non-smooth characteristics of wind speed series,it is difficult to achieve accurate prediction using a single prediction model.For this reason,a data decomposition algorithm is used to eliminate the original data noise and improve the data quality.An information theory-based approach is introduced to obtain the feature inputs of the prediction model and reduce the model input dimension.Finally,a Gaussian process regression prediction model based on the decomposition combination is established.The proposed method is applied to the actual wind speed data,and it is verified by simulation that the method can achieve high prediction accuracy.The model is applied to the wind speed data set including environmental variables to further demonstrate the reliability of the proposed model.(2)To reduce the impact of meteorological factor data redundancy and wind farm data noise on the accuracy of subsequent wind power prediction,a Gaussian process regression model based on mutual information feature extraction and wavelet threshold noise reduction is proposed to predict wind power time series with noise using a combination of historical wind power data and meteorological factor data.Mutual information is used to analyze the dependencies between meteorological factors and wind power time series and to select an effective set of input features.A wavelet threshold noise reduction method is used to separate the effective signal from the noisy signal in the wind power data.A Gaussian process regression prediction model is established to realize the wind power prediction.It is verified by real wind farm data that the proposed model can effectively improve the prediction accuracy in noisy scenarios.(3)The reliability of wind power prediction methods is an important factor to improve the security of power systems.Since meteorological factors affect the future values of wind power to some extent,a hybrid deep learning wind power prediction method is proposed.The method includes meteorological factor selection,data decomposition,feature extraction and construction of a hybrid prediction framework.First,a data preprocessing strategy based on maximum information coefficients,which is used to find the smallest input features and mine the abstract relationships hidden between sequences,and empirical wavelet transform,which adaptively decomposes the raw data to reduce the difficulty of prediction,is adopted.Then feature extraction is performed based on a one-dimensional convolutional neural network to obtain feature inputs for a set of input variables containing historical values of meteorological factors and wind power.Finally,a prediction model based on a two-way gated cyclic unit is constructed for predicting the future actual wind power.The proposed method is applied to the actual wind power data,and the simulation results show that the proposed method outperforms other comparative models.