摘要：With the advent of the era of “Big Data”, the application of the large-scale data is becoming popular. Efficiently using and analyzing the data has become an important problem. Traditional knowledge reduction algorithm read small data samples once into the computer main memory for reduction, but it is not suitable for large-scale data. This paper takes the large-scale sensor monitoring dynamic data as the research object and puts forward an incremental reduction algorithm based on Map-Reduce. Using Hash fast partitioning strategy this algorithm divides the initial data set into multiple subdatasets to compute which has greatly reduced the calculation time and space complexity of each node. Finally，through some experiments on the data sets in UCI machine learning repository based on Hadoop platform，the algorithm is proved more efficient and suitable for large-scale dynamic data. Compared to the traditional algorithm, the highest speedup of the parallel algorithm can be increased up to 1.55 times.
