Modeling And Performance Evaluation Method Of Wind Turbine Based On Improved LSTM

In recent years,with the continuous progress of wind power generation technology,wind turbines are gradually becoming large-scale and complex in structure,coupled with the relatively bad operating condition and the installation site is very remote,so that the operation performance evaluation of wind turbines is particularly important.Since the output power is a direct reflection of the performance of wind turbine,modeling and analysis based on operation data has become a hot topic for performance evaluation of wind turbine.Based on the data of supervisory control and data acquisition(SCADA)system,the wind turbine modeling and performance evaluation is realized.The main contents are as follows:(1)Considering the complexity of wind turbine operating conditions,the strong nonlinearity between monitoring variables,and the traditional machine learning algorithms is difficult to learn the characteristics of high dimension.The deep learning is introduced,and the "memory" function of long short-term memory neural network is used to construct wind turbine power prediction model based on the characteristics of wind turbine output power.In order to improve the prediction ability of the model,the attention mechanism is embedded between the hidden layers of the network to reasonably allocate the attention weights of different state characteristics,and the back propagation through time and dropout technology are introduced to solve the weight updating and over fitting problems caused by using the model,so as to further improve the prediction effect of the model.(2)Based on the statistical theory,the power residual probability density estimation model of nonparametric kernel density estimation is used to quantify the abnormal degree of wind turbines.The bandwidth value of nonparametric kernel density estimation model has a great impact on the fitting effect of the model.By using the least square cross validation method to obtain the optimal bandwidth value,the prediction accuracy of the model is improved.(3)In addition to wind turbines sudden situation,the performance of wind turbines usually changes through multiple decay stages,and finally reaches the failure or failure stage.According to the nonparametric kernel density estimation,the cumulative probability distribution function of the model is obtained.Power residuals are divided into different state intervals by significance level.Finally,a finite state space discrete Markov state transition model based on stochastic process is established to analyze the change trend of wind turbine operation state by calculating the transition probability between different states and combining with the abnormal index.It makes it more intuitive and effective to detect the changes of the operation behavior state of the wind turbine,make an effective evaluation of the performance of the wind turbine,and formulate the best maintenance strategy priority.It has a certain reference value to improve the operation and maintenance efficiency and economic benefits of the whole wind farm.