Modeling Wind Turbines In Mountain Based On Machine Learning

The operation control and fault diagnosis of wind turbines require wind turbine models with high accuracy.Due to the variable terrain and meteorological conditions,the existing operation characteristic of wind turbine model is no longer suitable for mountain wind turbines.Therefore,it is particularly urgent to model mountain wind turbines considering the actual environment.Based on the measured data,the machine learning is used to model the operating characteristics of mountain wind turbine.This article contains the following sections:1.Combining data cleaning method with data balancing method preprocessed the measured data.Firstly,the common anomaly data cleaning methods were analyzed,and the data density-based cleaning method was used to achieve the cleaning of abnormal data.Then,the cleaned data was clustered based on the DBSCAN algorithm,to analyze the distribution of data samples in various wind conditions,and the integrated sampling method of ADASYN+ENN was used to solve the problem of uneven distribution of data samples after cleaning.Finally,the feasibility of the data preprocessing method proposed in this paper,which combines the data cleaning method and data equalization method proposed,was verified by experiments.2.In order to quickly obtain the overall operating conditions of multiple wind turbines in mountain wind farms,an efficient model for the power curve and power prediction of mountain wind turbines based on SVM was established.Without considering the uneven distribution of the data of each wind condition,the principle of the SVM algorithm and the analysis of its applicability to modeling the operating characteristics of mountain wind turbines were first studied,and an efficient modeling framework of power curve and power prediction of mountain wind turbines based on SVM algorithm was proposed.Next,because of the influence of turbulence intensity on the power curve,data mining and SVM algorithm were used to establish an efficient model of the power curve of mountain wind turbines.After the training set sampled and reconstructed,a power prediction model was established by SVM algorithm.Finally,based on above experiments,it is verified that the proposed power curve model and power prediction model based on SVM algorithm improved the modeling efficiency while ensuring certain accuracy.3.In order to evaluate the operating status of mountain wind turbines precisely,an accurate model of the power curve and power and speed prediction of mountain wind turbines based on ensemble learning was established.Based on the data of the balanced distribution of each wind state after data preprocessing,the principle of ensemble learning was first studied.By analyzing the applicability of the common ensemble learning architecture and individual learner in modeling the operating characteristics of mountain wind turbines,a precise modeling framework based on Stacking ensemble learning architecture was proposed for power curve and state quantity prediction.Next,based on the Stacking ensemble learning framework,an accurate model of the mountain wind turbine power curve was established.Then,considering the influence of the fluctuation of meteorological conditions,the input of the model was decomposed by the Hilbert-Huang transform,and an accurate model for power and speed prediction of mountain wind turbines was established based on the Stacking ensemble learning architecture.Finally,through experiments,the MAPE of the mountain wind turbine power curve model based on the Stacking integrated learning architecture was only 1.48%,and the MAPE of the power and speed prediction were only 0.59%and 0.25%,respectively.It is verified that the power curve and power and speed prediction model of mountain wind turbine have the characteristics of small error and high accuracy.