Enhancing evacuation efficiency is of paramount importance in the field of evacuation systems research. Evacuation systems often present observability limitations, and any abnormal observation of pedestrian density at the exits can diminish the effectiveness of evacuation control. Therefore, correcting the abnormal observation information at exits becomes imperative for improving evacuation performance. To address this issue, an algorithm based on the extended Kalman filter is proposed to predict pedestrian density, and a correlation mapping between normal and abnormal pedestrian densities is established. The algorithm incorporates a neural network fitting method to identify the parameters of the state and observation functions in the extended Kalman filter algorithm, enhancing the accuracy of system modeling by approximating nonlinearity. Moreover, an iterative update mechanism utilizing the error covariance matrix allows for fast prediction and correction of pedestrian density. Additionally, the algorithm incorporates a density control algorithm to formulate a pedestrian flow evacuation control strategy for abnormal evacuation scenarios. Comparative simulations are conducted by using the evacuation model in abnormal evacuation scenarios to evaluate the effectiveness of the proposed algorithm. The results show that, compared to the evacuation control strategy without data correction, the proposed algorithm achieves efficiency improvements of up to 38. 9% and 23. 26% in abnormal evacuation simulation and human-controllable scenarios, respectively, which provides an effective solution approach for control strategies in abnormal evacuation scenarios.