Research On Wind Blade Icing Prediction Based On LightGBM, XGBoost, And ERT Hybrid Models

In recent years,with the rapid development of information technology and communication technologies such as the Internet,the Internet of Things,and sensors,the rapid increase in data volume has become a serious challenge and a valuable opportunity faced by many industries.The progress of information technology in industry and the development and popularization of modern management concepts have led companies to rely more and more on information technology.In the industry at this stage,a large amount of equipment operating condition data has been stored,showing many characteristics of big data,but most companies have not dug out the value that these data should have.Nowadays,advanced information and communication technologies are constantly being incorporated into industrial equipment and continuously updated to promote the development of industrial equipment in the direction of automation,digitization,and intelligence.The Chinese government has put forward the concept of "Industrial Big Data" based on the above trends.Specifically,equipment and production lines are equipped with various sensors to capture data,then connect to the Internet through wireless communication,transmit and store data,and perform point-to-point real-time status monitoring of equipment operations or production processes.During the operation of the equipment,the external environment or unexpected events may cause certain changes in the performance of the equipment and cause failures.Sensing data can now be sensed through sensory technology,and the fault diagnosis of the equipment can be realized by accurately perceiving various factors in the operation of the equipment and using statistical learning knowledge.In order to make full use of the mass data in the industry and ensure the accuracy and efficiency of diagnosis,this paper will use a statistically driven learning model to predict the icing conditions of fan blades.This paper divides the fan operating status data into different time windows,uses the models LightGBM,XGBoost,and ERT models for nested fusion to obtain a hybrid model,which reduces the scope of suspicious fault data and guarantees almost complete coverage under more accurate conditions.The fault data is better under the subdivided time window.This article describes the model LightGBM and the connection between LightGBM and XGBoost,introduces feature engineering based on business context,and uses LightGBM,XGBoost,and ERT models based on different time windows to gradually narrow down the scope of suspicious fault data,conduct training and testing,then Compare this model to improve the prediction.