Wind Turbine Generated Production Prediction Based On Wind Speed Prediction And Health State Evaluation

Forecasting wind power is the basis for effective wind power system dispatching and planning.Increasing the prediction accuracy can reduce the jeopardization of wind power integration to the stability and economy of the power system.On the one hand,as wind speed is the most critical factor affecting the power generation of wind turbines,precise wind speed prediction is fundamental and essential for short-term wind power forecasting.On the other hand,as for the middle-term and long-term wind power generation forecasting,previous forecasting methods have not paid sufficient attention to the fact that the health status of wind turbines would impact their functions.This article focuses on the issues of short-term wind speed forecasting and long-term wind turbine’s power generation forecasting.(1)A hybrid wind speed prediction model based on time series decomposition,feature selection,and basic prediction model with the framework of synchronous optimization is designed to ward off the challenge of strong non-stationary in short-term wind speed prediction.For starters,the raw wind speed sequence is decomposed into several subsequences by Variational Mode Decomposition,and the sub-sequences are denoised by Singular Spectrum Analysis.Then,using Long Short-Term Memory Networks to predict the values of the sub-sequences,and then integrating them to obtain predicted wind speed values.The sequence model simultaneously optimizes the critical parameters of the model.The experiment results show that the proposed hybrid model has the best multi-step prediction performance compared to traditional models.(2)A method for evaluating the health status of the wind turbine based on multivariate observations Hidden Markov Model(HMM)is proposed,based on which we conduct the power generation prediction with the impact of health status considered.Firstly,we categorize the health status of the wind turbine into five levels,and the performance observation matrix is obtained by preprocessing the multivariate observations data.Combined weights are introduced to describe the correlation between different performance observation indexes,and the performance observation matrix is trained with the improved Baulm-Weltch algorithm so that we could obtain the parameters of HMM.Then,the Viterbi algorithm is utilized to predict the health status sequence;meanwhile,the forward-backward algorithm is used to predict the probabilities of five health status at a specific moment.(3)A long-term wind turbine power generation prediction method based on the wind speed prediction and health status evaluation is proposed.Firstly,the cleaned wind speedpower data is used to establish the wind power curve(WTPC)of the unit at various levels of health.Then the health status evaluation results obtained by the HMM are brought into the WTPC to construct the dynamic WTPC of the wind turbine considering the health impact.Then the wind speed prediction results are fitted with Weibull distribution function so that we can obtain the wind speed distribution probability density function.Finally,the wind turbine generating capacity can be predicted by the wind speed probability density model and the dynamic WTPC.The validity of the proposed method is verified through a case of the SL2000 / 100 wind turbine.The wind speed time series studied in this paper has long-term trends,cyclical fluctuations,and random fluctuations,which makes this study also suitable for the application of time series prediction algorithms in other industries.Plus,the health assessment model proposed in this paper is not only applicable to the degradation process of wind turbines,but also to other equipment with stochastic degradation.