Abstract: Prediction of Lindz-Donawitz gas holder level in iron and steel enterprises is difficult to model. A Lindz-Donawitz gas holder level prediction PNN-HP(2)-ENN model is proposed, which combines the properties of probabilistic neural network, HP(Hodric-Prescott) filter, and Elman neural network. The simulation uses data from an iron and steel enterprise, and the results show that the predictive effect of the function is excellent, with high classification accuracy and less time used. Results further show that it is more suitable for blast furnace gas output prediction than other methods, and can provide some operating proposals for the reasonable scheduling of byproduct gas.