基于Spark的模糊时间序列预测模型研究-计算机技术专业论文.docxVIP

AbstractIn Abstract In recent years，with the development of electronic devices，sensors and storage service，a large amount of dam has been accumulated in many domains such aS finance and medicine，leading to various requirements of data process application，and big data technology has attracted widespread attention from people from all walks of life．As stock market thrives，the scale of stock time—series data grows continually,not only in the length of series，but also in the number of series．Therefore，it is difficult for people to analyze，understand and process． For the prediction of financial time—series data，the model based on fuzzy time series and the genetic algorithm Can forecast several stock indexes with a high degree of accuracy．Fuzzy Logic applies explicit fuzzy rule definition to direct its prediction， with the result that output varies more smoothly when input is changed sharply． Genetic Algorithm is used to search the reasonable fuzzy set boundaries，improving the quality of solutions by evolutionary operators until convergence is seen．However, to our knowledge，there is hardly any research about models of fuzzy time series on Spark，which is large—scale data processing engine．In the face of arrival of the age of big data，we urgently need to integrate fuzzy time series forecasting method into cluster computing． The core work of this paper contains the following three parts qlb begin with，since the most time·consuming portion of the prediction model is in genetic algorithmj when time—series data i s massive，it is inefficient to the algorithm it．In this paper,three versions of parallel genetic algorithm based Spark were described Secondly,the above three parallel genetic algorithm models implemented on Spark，and models tested on small—scale and large-scale SAT problem benchmark separately to do comparison with the difference in convergence，optimal solution， running time and scalability Finally，parallel single—series prediction model and parallel multi—series pre