Research And Application Of Sea Level Wind Speed Anomaly Detection Based On LSTM-VAE Model

Green low-carbon provides new momentum for high-quality development.Among all renewable energy,wind energy is the most promising one.To a certain extent,wind power generation ensures the overall stability of national energy supply and demand.At the same time,wind power has a large fluctuation due to wind speed.Therefore,accurate wind speed forecasting is of great significance for improving wind power forecasting,optimizing grid dispatch,and even green sustainable development.Aiming at the problem that the existing sea level wind speed prediction model lacks consideration of the intermittent and nonlinear characteristics of wind speed,which leads to the low accuracy of short-term wind speed prediction,a sea level wind speed anomaly prediction model is proposed by integrating LSTM(Long Short-Term Memory,LSTM)and VAE(Variational Autoencoders,VAE)models(denoted as LSTM-VAE).The main research content and work of this article are as follows:(1)Dataset construction and pre-processing.It Involved the original data of the European Medium Weather Forecast(ECMWF)and the pre-processing of the actual wind speed data of the sea flat plane in Zhangzhou City.Firstly,because of the limited distribution area of sea level meteorological stations in the research data set(Zhangzhou City,Fujian Province)and the low spatial resolution of the data provided by ECMWF,this paper uses the experimental verification method to select the inverse distance weight interpolation method to interpolate the original ECMWF data.;secondly,aiming at the problem of missing values and outliers in the sea level wind speed data obtained from ECMWF data and sea surface wind meteorological observation stations,the missing value repair based on mean filling and the outlier processing based on local outlier factor are carried out.Finally,to construct the prediction model of sea level wind speed anomaly,the construction of meteorological factors is studied,and the construction method of meteorological factors based on geometric method and gradient transformation is proposed.(2)Construction and performance assessment of sea level wind speed abnormal detection model.Aiming at the obvious time series and periodicity of sea level wind data,a sea level wind speed anomaly prediction model combining LSTM and VAE is proposed.Firstly,the LSTM model is used to train the historical sea level wind speed time series,and the appropriate step size is selected by the experiment.Secondly,since the LSTM model cannot capture the structure and characteristics of sea level wind speed data well,it is proposed to input the features obtained by LSTM training and the meteorological factors constructed from ECMWF into the VAE model to extract lowdimensional implicit variables.Finally,the recall rate,precision rate,accuracy rate,f1 value,and ROC curve are used as evaluation indexes to carry out sufficient model validity verification experiments from two perspectives.One is that compared with the traditional regression model and the common anomaly detection model,the proposed model is verified to have good anomaly detection accuracy.The f1 value index is 0.87,which is 0.07 higher than the best data of the comparison method.Secondly,the ablation experiment was carried out by the control variable method to verify the validity of the model combination method,the combination order,and the validity of the model.(3)Design and implementation of sea level wind speed anomaly detection system.In this paper,the LSTM-VAE model is used as the core to design and develop the sea level wind speed anomaly detection system in Zhangzhou City.According to the software system development process,the feasibility analysis,demand analysis,and architecture design are carried out,and the system is designed in detail and the function is realized.The system further verifies that the proposed sea surface wind speed anomaly prediction model combining LSTM and VAE has a certain business application value.