The Rainfall Predication Research Based On Storm And Online Sequential Extreme Learning Machine

Rainfall is an important parameter of disaster prevention and mitigation,and it reflects the trend of disaster occurrence.The rainfall has an important influence on agricultural production,soil and water flow and engineering application.It is helpful to forecast the rainfall in one area and help agriculture,Water sector to improve the ability to prevent droughts and droughts,the harm will be reduced to a minimum.With the recent years,China's floods continue to frequent,how to accurately and timely use of weather data on rainfall forecast has become increasingly important.The advent of the era of large data,but also to the weather forecast industry has brought new challenges.Meteorological data are mainly from ground observation,meteorological satellite remote sensing,weather radar and numerical forecast products.These four types of data accounted for more than 90% of the total data,directly applied to meteorological services,weather forecasts,climate prediction and weather services.The stream data is a set of digitally encoded and continuous signals.Under normal circumstances,the data stream can be regarded as a continuous extension over time and unlimited widely used in network public opinion analysis,stock market trends,satellite positioning,real-time financial monitoring,Internet of Things monitoring and real-time weather monitoring and other fields.In the field of rainfall forecasting based on large-scale meteorological data,there is still a lot of room for development.For the traditional rainfall forecast,the use of offline weather data,the use of machine learning methods for mass training,that is,all the training samples after a onetime study,the learning process is no longer continue.However,in practice,all samples of training sample space can not be obtained at one time,but often in chronological order.Although large-scale clustering can alleviate a large amount of data to a certain extent,the problem of insufficient computing capacity,but for the newly arrived data,but can not quickly learn to learn and update the learning knowledge obtained.Aiming at the problem of real-time calculation and mass processing of meteorological data,this paper presents a rainfall Predication Model Based on Storm and Online Sequential Extreme Learning MachineThe main contents and innovations of this paper are as follows:(1)The off-line batch forecasting method for meteorological data can not reflect the change of rainfall in time,and a prediction model of rainfall based on online sequence extreme learning machine is proposed.Aiming at the large-scale and real-time characteristics of meteorological data,the online learning of the extreme learning machine is optimized.The model first initializes a number of online extreme learning machine models.When the new batch data is continuously reached,the model can continue to study the new samples on the basis of the existing training results and introduce the random gradient descent method and the error weight adjustment The error of the new prediction results error feedback,real-time update the error weight parameters to improve the model prediction accuracy.(2)In the data preprocessing stage,the correlation coefficient between the decision attributes is used to analyze the meteorological data,and the correlation coefficient is used to filter the prediction attributes,which reduces the complexity of the meteorological data.Model training efficiency.In addition,the Storm flow large data processing framework combined with Kafka distributed message queue,the large-scale meteorological data for parallel training.Experimental results show that the algorithm is running on the Storm platform,with excellent parallel performance and prediction accuracy.