Study On The B-P Neural Networks Of PM₁₀ Concentration Forecast Of The Four Seasons In Xi'an City

The forecast model of PM₁₀ concentration of the four seasons in Xi'an city has been studied according to the PM₁₀ concentration date daily and the ground meteorological element data daily that monitored from February of 2001 to march of 2006.Analysed the PM₁₀ concentration's and the meteorological element data's change anaualy,seasonly, the obviously change of the PM₁₀ and the meteorlogical element seasonly has been found. The significant correlation between the PM₁₀ concentration and other meteorological elements is analysed by use of the paitical correlation analysis before building the B-P neural networks model. Tranquilization test of source release is condected, the optimal selection for the modeling modules is made and indicated the four season modules is valid.The change of the PM₁₀ concentration and the meteorlogical elements is the characteristic of non-linearity. B-P neural networks model can demonstrate the discipline between the change of air pollution concentration and the meteorologic data which effects air pollutant concentration due to the ability of non-linearity. B-P neural networks model is established according to the PM₁₀ concentration of four seasons in Xi'an city. Owing to such storage as plunging part minimum easily, slow convergent rate and so on, the atructure and the algorithm of B-P network are improved. A improving algorithm is adapted to reform the capacity of neural network. The trying means is adapted to confirm the number of neural network. By means of the simulation and practice, the precision of spring model is 80.81%, summer's is 81.86%, autumn's is 79.98%, and winter's is 82.03%. relative erro and absolute erro is -0.03mg/m³ ～0.03mg/m³ and-30%～30% respectively. This B-P neural networks model is further feasible and further suitability according to the erro analysis and the precition analysis.