Abstract: Real-time and accurate state monitoring is essential for vehicular power batteries, which depends on the information data collected by a large number of sensors. In long-term use, high-frequency vibration and loose connectors make local sensors fail, which leads to abnormal data collection. Considering the fact that researches on the data missing and stagnation of the data update are few, a method of monitoring and correcting abnormal data based on the bi-directional long short-term memory network and the least square support vector regression is proposed. Thevenin model is used in the modeling, and the recursive least square method is used in parameter identification. The simultaneous input and state estimation (SISE) algorithm is used in the battery states estimation. In the experiment, the estimation error of the proposed method is kept at about 5% under six cases of mixed abnormal test conditions, therefore it is proved to be effective.