Abstract: The three-way catalyst（TWC） has certain oxygen storage and release function,which directly affects the transformation efficiency of pollutants in the volatility of the catalyst in the excess air coefficient（φ_a） fluctuation.Therefore,the monitoring and prediction of the oxygen storage and release process of TWC can help to decrease and control the transient emissions of the engine.However,the oxygen storage and release process of TWC is part of its complex chemical reaction and lacks the way to direct observation,and modeling based on detailed chemical mechanisms is too complicated to meet the needs of real-time control.Therefore,this study is based on the chemical reaction mechanism modeling,using the long and short-term memory（LSTM） neural network to observe time series data,and constructs a oxygen storage observation model,which accurately and quickly feeds back the TWC oxygen storage and downstream φ_a.The results show that the observation model has an average relative error of 5.87% for the prediction results of the vehicle’s oxygen storage under different operating conditions,and the average relative error of the downstream φ_a prediction results is about 0.27%,and the prediction time is about 0.77%of the mechanism model.