Research On Short-term Wind Power Forecasting Based On Multivariate Optimization ELM And Error Correctio

To achieve the goal of "carbon peak,carbon neutral" as soon as possible,China is now actively carrying out innovation and breakthrough research on new energy technologies.Wind power generation has become one of the world’s largest new energy generation technologies due to its sound theoretical system and low power generation cost.Our goal in this study is to increase the reliability of wind power’s short-term forecasts.We approach the research from a variety of angles,with the following primary research findings:An improved complete ensemble empirical mode decomposition algorithm with adaptive noise(ICEEMDAN)is presented to decompose the initial power data into a number of component sequences to solve the problem of the original wind power’s low prediction performance due to its own significant volatility,and combine the sample entropy(SE)principle to evaluate the complexity of the component sequences and reorganize the sequences of similar complexity by superposition.The initial wind power series’ volatility is significantly reduced after preprocessing with the ICEEMDAN-SE approach,and the ability to analyze power series is improved while maintaining efficiency,which prepares the conditions for the subsequent improvement of the prediction accuracy.When predicting wind power sequences with various features,the kernel extreme learning machine(KELM)with an unique kernel function is unable to strike a balance between performance and stability.A combined kernel extreme learning machine(CKELM)that mixes the Poly kernel function and RBF kernel function is suggested in response to the aforementioned issues.The prediction performance of CKELM is dependent on the parameter settings.Based on the scientific modification of the kernel parameters,CKELM can adjust its model structure for sequences with different characteristics to show different prediction features,thus improving the prediction accuracy.The CKELM model kernel parameters are difficult to set,and the improved chaotic sparrow search algorithm(ICSSA)is introduced to seek the kernel parameters of CKELM.Compared with the traditional sparrow search algorithm(SSA),to boost population diversity,ICSSA introduces the Tent chaotic mapping function.It also combines dynamic inertia weights,backward learning theory,and adaptive t-variation strategy to optimize the update strategy of individual positions.These improvements strengthen the algorithm’s optimization-seeking accuracy and convergence speed,which in turn enhances the prediction model’s capacity for learning and generalization.To further improve the short-term prediction accuracy of wind power,based on the theory of spatial correlation of wind power,the TCCA model is proposed to correct for the wind power prediction error by combining the temporal convolutional network(TCN)and the efficient channel attention(ECA)mechanism.The preliminary prediction power and preliminary prediction error are first obtained by preliminary prediction,and the correlation is measured by utilizing the maximum mutual information coefficient method for other wind turbines around the target wind turbine,and wind turbine powers with higher correlation with the target wind turbine power are selected,and the preliminary prediction error and the screened other wind turbine powers are input into the TCCA model together for prediction to obtain the corrected error.The final predicted power is obtained by superimposing the initial estimated power with the corrected error.To enhance the prediction accuracy of wind power,a combination model of ICEEMDAN-ICSSA-CKELM-TCCA for wind power forecasting is suggested in this paper which is based on signal decomposition,optimization algorithm,neural network,and error modification.The results of multiple experiments with multiple datasets from a wind farm in Yunnan Province,China,indicates that this model has high forecasting accuracy and generalizability for wind power.