To address the impact pressure sensor measures the temperature drift problem, the development of a rail car for high-performance sensor compressed air system pressure measurements. This paper presents new piezoresistive pressure sensor temperature compensation model for combining GA-PSO algorithm to optimize BP neural network, the feasibility of this model by MATLAB simulation. Simulation results show that: compared with the BP-LM algorithm and GA-BP algorithm, a new optimization algorithm converges faster 89%, the absolute value of prediction errors were 0.025kpa less, and small-scale changes and meet the "fast, accurate , steady high performance requirements.Plant site running well, compressed air system on the same day in different periods railcar running pressure measurements were about 0.75Mpa.