Research On Adaptive Wind Power Prediction Metho

With the innovation of technology and the improvement of social productivity,the problems of energy depletion and environmental pollution have become increasingly serious.Wind energy,with the characteristics of wide distribution and renewable,has become an important part of clean energy.Reliable wind power forecasting is the key to performing optimal scheduling of wind energy.However,it is challenging to obtain accurate wind power forecasting due to inherent intermittency and randomness of wind energy.In addition,as time passes,the characteristics of wind power data are bound to change,which directly leads to the degradation of prediction model performance.In order to enhance the precision and reliability of wind power prediction,this paper investigates the research on adaptive ultra-short-term prediction methods.The main research focuses of this study include:(1)For traditional deep learning-based wind power prediction methods,the models are difficult to update and unable to dynamically learn information from new samples.Therefore,the performance of the model will inevitably deteriorate over time.To address this problem,a wind power prediction method which incorporating deep learning and adaptive modeling mechanisms is proposed.This method introduces local weighted learning and adaptive error correction mechanisms,and integrates the big data mining capability of deep learning models with the adaptive updating ability of local weighted learning models through online adaptive ensemble,which effectively improving the prediction accuracy.Finally,the effectiveness of the method is verified by real wind farm data.The effectiveness of the proposed method is validated using real wind farm data.(2)Traditional ensemble-based wind power prediction methods with the defects of insufficient adaptability,the structure of individual models is homogenous,and neglect of integrated pruning.To address these problems,an adaptive wind power forecasting method based on selective ensemble of offline global and online local learning is proposed.Firstly,diverse based models are generated by multi-modal perturbation mechanism.Subsequently,in order to ensure the effectiveness of the ensemble,diversity and accuracy objective functions were defined,and NSGA-II was used to implement ensemble pruning.Furthermore,the adaptive fusion of the individual models is achieved according to the Bayesian rule.The validity of the method is verified by an actual wind power data in Belgium.(3)For the ensemble wind power method based on just-in-time(JITL),the diversity of individual model is limited which leads to unsatisfactory ensemble performance and high-frequency individual model construction and discarding lead to a large consumption of computational resources.To address these problems,an online selective ensemble adaptive wind power prediction method based on process state identification is proposed.The method introduces a multimodal perturbation mechanism and hypothesis testing to achieve online selection and adaptive integration of forecasting models.In order to reduce the reconstruction frequency of the model,process state identification method is employed to realize the efficient update of the model library.Finally,the effectiveness of the proposed method is verified by an actual wind power data.For traditional ensemble immediate learning methods for wind power prediction,the insufficient diversity of the just-in-time(JIT)base models while frequent construction and discarding of models result in substantial computational resource consumption.To address these issues,this paper proposes an online selective ensemble adaptive wind power prediction method based on process state recognition.This method introduces a multi-modal perturbation mechanism and statistical hypothesis testing to achieve online selection and adaptive integration of individual JIT models.To reduce the frequency of model reconstruction,a process state recognition method is introduced to enable efficient updates of the base model library.Finally,the effectiveness of the proposed method is validated using actual wind power data.