Research And Implementation Of Maximum Wind Speed Prediction System Based On Machine Learning

With the rapid development of society,the demand for energy is also increasing.The problems of air pollution,greenhouse effect and energy shortage caused by fossil energy consumption have to be paid attention to.In addition,China strives to achieve the goal of carbon neutrality by 2060,and puts forward higher requirements for the use of clean energy.Wind is a clean and renewable energy source,and accurate prediction of wind speed is of great significance for improving the efficiency of wind power generation,guiding agricultural production and ensuring large-scale competitions.In this paper,the historical meteorological data at three-hour intervals of 24 observation stations in a meteorological station in 2017 were analyzed,the characteristics and distribution of meteorological data were preliminarily analyzed,and the maximum wind speed prediction was carried out.The main contents and achievements of this paper are as follows:(1)Conduct pre-processing work such as sorting out,classifying and cleaning the original meteorological data.Multiple machine learning algorithms such as Linear Regression(LR),Support Vector Machine(SVM),Random Forest(RF),XGBoost and Light GBM are used to establish preliminary prediction models on the data of 24 monitoring stations.Compared with the model prediction results,it is found that the prediction effect of integrated learning algorithm RF,XGBoost and Light GBM models is better than that of single machine learning model in multiple model evaluation metrics.(2)Research on improving the accuracy of prediction model.A model merging method for replacing and combining base learners is proposed.Using Stacking model merging method,the algorithm with good prediction results,such as SVM,RF,XGBoost and Light GBM,is used as the basic algorithm input of Stacking.Then,the LR algorithm is used to dynamically adjust the weights between the results of each basic learning algorithm to improve the prediction accuracy of the model.In terms of algorithm application,jpmml(Java Evaluator API for Predictive Model Markup Language)is used to transform the trained model across platforms,so that it can be called in the back-end system.(3)Research on the application of maximum wind speed prediction system for meteorological field.First of all,in view of the cumbersome operation and build of the maximum wind speed prediction system,a container-based development and operation environment is proposed.Compared with the traditional way of building,it greatly improves its portability,saves the template of our node environment configuration,and improves the utilization of hardware resources and the isolation between applications.Secondly,for the stability and high availability of the maximum wind speed prediction system,the commonly used message middleware Kafka and the coordination service framework Zookeeper in the industry are used to decouple the system and ensure the stability and reliability of the prediction system when receiving weather data.Use Heartbeat and Zookeeper to ensure high availability of each module.Thirdly,in view of the data storage of the maximum wind speed prediction system,due to the continuity of data generation and the particularity of data format,the single machine storage of the traditional relational database cannot meet the future data storage.It is proposed to deploy the big data framework Hadoop and the non-relational database HBase in the container environment to store meteorological data.Finally,the maximum wind speed prediction model is implemented to meet the needs of the field,and it is approved by users after preliminary application.