In practical applications, the locating accuracy of Mach Zehnder fiber sensing systems is prone to the initial phase shift, thus a new type of Genetic Algorithm method based on dual Mach Zehnder signal correlation coefficient is proposed for real time control of input polarization, which can keep the signals’ correlation coefficient stable. When the correlation of the detection signals is degenerating, it can be recovered to be steady by changing the input polarization state. Test results show that, the genetic algorithm based on dual Mach Zehnder signal correlation coefficient can make a fast search for the optimal state of polarization, and maintain a steady signal correlation, and the system stability is significantly improved.