Research On Hybrid Short-Term Wind Power Forecasting Based On Correntropy Long Short-Term Memory Neural Network

With the advancement of the energy revolution,renewable energy has attracted worldwide attention due to its green,low-carbon,and non-polluting characteristics.Among them,wind energy has become one of the most potential alternative energy due to its clean and widely distributed characteristics.However,wind energy has inherent characteristics such as intermittency,randomness and volatility,which severely restrict its utilization efficiency.Developing a predictive model with high precision and strong robustness is one of the effective measures to improve wind energy utilization efficiency.Up to now,most wind power prediction models are designed based on the mean square error(MSE),which is very suitable for the case that the prediction error obeys Gaussian distribution.However,the actual wind farm is affected by various random factors such as meteorological conditions,wind turbine operating conditions,data collection error,etc.,which makes it have a large number of outliers.Meanwhile,the strong nonlinear process of converting wind energy to wind power will also change the power distribution characteristics of wind power.Both of these two factors will cause the wind power series to not completely obey the Gaussian distribution.Therefore,this paper utilizes the maximum correntropy loss(MCC)and mixture correntropy loss(MMCC)that are not sensitive to outliers and non-Gaussian distributions to replace the MSE loss of the classic Long short-term memory network(LSTM)to develop two novel robust predictors.Subsequently,combined with the data preprocessing strategy,two hybrid forecasting models are proposed.The specific research work is as follows:1)To better extract the local features of wind power series,the variational mode decomposition technology(VMD)is used to smooth the original wind power series.Meanwhile,in view of the defect that VMD technology requires a preset parameter K,an adaptive variational mode decomposition(IVMD)algorithm is developed via using the maximum correntropy criterion(MCC).Subsequently,in order to improve the prediction efficiency and strengthen the correlation between the sub-modes,the sample entropy(SE)function is employed to reconstruct the sub-modes to construct the IVMD-SE data preprocessing model;2)Considering that the wind power series doesn’t completely obey the Gaussian distribution,the maximum correntropy loss is used to replace the MSE loss of the traditional Long short-term memory neural network(LSTM)to develop maximum correntropy Long short-term memory neural network predictor(MCC-LSTM).Then,the IVMD-SE-MCC-LSTM hybrid forecasting model is designed by combining with IVMD-SE data preprocessing model;3)Taking into account the data complexity and volatility of a single wind turbine,and to improve the flexibility of MCC loss,the introduction of maximum mixture correntropy loss(MMCC)with Laplace kernel function and Gaussian kernel function to replaces LSTM’s MSE loss.Then,the particle swarm optimization(PSO)algorithm is adopted to optimize the mixture coefficients and the kernel parameters to construct a mixture correntropy Long short-term memory network predictor(PMC-LSTM).Finally,combined with the data preprocessing strategy,the IVMD-SE-PMC-LSTM hybrid prediction model is proposed;4)Different datasets are employed to verify the developed prediction method and compared with other existing prediction methods.