Research On Meteorological Data Prediction Algorithm Based On Improved Bayesian Network

With the development of computer science and Internet technology, people’s production and living is becoming more and more convenient, and the amount of data is growing with it. The big data contains a wealth of knowledge and a lot of rules. If we can find out the useful information, it will have a great help to our future life. Especially in the field of meteorology, the help will be more obvious. For a long time, the weather forecast plays an important role in people’s production and living. From takeoff and rocket to planting and dressing, weather forecast profoundly affects our country’s scientific research, economic construction and people’s life. In recent years, the modernization level of meteorological service and the modern meteorological system are constantly improving. With it some new types of data emerges from the ground, satellite observations and numerical forecast, and the data volume has reached PB level. Unfortunately, although the data is growing very quickly, the method of processing data is relative backward. We have a lot of difficulties to process and calculate the mass data using the traditional data mining methods.In this context, the emergence of Hadoop makes it possible to mine the mass meteorological data efficiently. The core idea of Hadoop is unified management and scheduling for computing resources, and forming a resource pool to provide on-demand services to users. The basic idea of mass meteorological data mining based on Hadoop is putting the traditional meteorological data mining algorithm and the Hadoop platform together. So we can make full use of the mass data by the super computing power of the cloud platform and obtain better weather prediction finally. As one of the most commonly methods in Hadoop software platform, Hadoop has high fault tolerance,high throughput, low cost and many other advantages.This dissertation deeply studies the data mining method based on Hadoop platform and the characteristics of meteorological data. Considering some deficiencies of the traditional Bayesian network classification data mining methods, and combining with the Hadoop cloud platform to deal with the advantage of the huge amounts of data, this dissertation puts forward the improved algorithm based on Bayesian network classification of Map Reduce. The dissertation mainly do the following research:(1) Considering the large-scale characteristic of meteorological data, this dissertation uses the Hadoop platform to deal with the meteorological original data set, and calculate the correlation coefficient between decision attribute and the other attributes. It uses the correlation analysis technology to select prediction attributes. To a certain extent, it reduces the complexity of the model training.(2) This dissertation analyzes the superiority and inferiority of some typical meteorological data mining classification algorithm. Based on the correlation characteristics of meteorological data, this dissertation chose the bayesian network classification algorithm and use maximum information coefficient learning the bayesian network structure. It is proposed to solve the uncertainty and relevance of things.(3) In bayesian classification model training process, this dissertation adopted the accuracy evaluation pattern. If the Classification model does not meet the accuracy requirement, it may need to continually modify parameters, and obtain the optimal classification model. The experimental results show that the improved algorithm is better than the existing algorithms in both computational efficiency and performance.