Based on temperature forecast values of county stations in Ningxia in 2016, the dynamic optimal PP method of township temperature forecast was established using dynamic, training optimal and PP method. The tests show that the quality score of PP method was significantly better than that of Ningxia township business forecast. The mimimum and maximum temperatures increased by 4.74% and 8.20%, respectively. The appearance frequency of optimal sample length selected by the dynamic optimal method was not exactly the same in different counties and in different months.Therefore, it was appropriate that the dynamic optimization techniques were used to select the optimal sample length.In one year, the accumulated frequency of optimal sample length decreased logarithmically with increasing number of samples, among them, 3, 5, 7, 11, 19, 22, etc. were easily selected as the optimal sample length. The optimal sample length reflected indirectly that the relationship between county temperature and township temperature was periodic.