Research And Design Of The Pulsar Candidate Feature Selection Algorithm

With built of Five-hundred-meter Aperture Spherical radio Telescope(FAST)and the debugging operation of 19-beam receiver,the amount of pulsar sky survey data has entered the PB level,massive growth in the number of candidates generated by pulsar search.In massive pulsar candidate information,a large number of interference signals brings great difficulties to candidate recognition.The open source software package PRESTO is the main search software of FAST.The sifting algorithm for periodic signal screening is programmed according to the experience of previous researchers.There are many problems in the algorithm.Therefore,in this paper,the existing sifting algorithm in pulsar search data processing is studied and improved,and a new filtering algorithm is designed.Firstly,this paper studies the PRESTO software package and the k-means algorithm used in the project.Subsequently,the sifting algorithm which used to screen periodic signals is studied,and the problems in the selection of pulsar candidate signals are pointed out.Finally,to improve the sifting algorithm.On the one hand,the program for obtaining the DM value is optimized,improving the run time of the algorithm;on the other hand,K-means algorithm is introduced to improve the existing problems.In the original algorithm,the method of selecting eigenvalues based on experience is changed to the method of clustering.Modify k-means algorithm to integrate it into sifting algorithm,and extract periodic signal features as data sets into k-means algorithm.Then the clustered data are added to the periodic signals to construct a new sifting algorithm.The final experimental data show that the improved algorithm improves the accuracy of screening.