Abstract: The factors that influence the reminded oxygen consistence in PTA process are introduced. In order to filter the noise of the data, a new method based on wavelet transformation theory is presented in this paper. The BP network is presented and improved. Simulation is made with or without wavelet noise filtering. The results of the two situations are compared and it indicates that the predictive precision and the generalizing ability of the neural network based on wavelet noise filtering are better. The forecast of abnormal state is made by the calculated reminded oxygen consistence. The comparison shows that the forecast of abnormal state can be made rightly.