Abstract: For boiler state monitoring in complex environment, a single sensor often fails to provide comprehensive and accurate information. A monitoring method based on multisensor fusion was proposed, which enhanced system performance by integrating data from different types of sensors. The principles of system construction and the theories and optimization strategies of data fusion algorithms, such as Kalman filtering and KalmanNet, were analyzed. Results show that multisensor fusion significantly reduces prediction errors, with Kalman filtering reducing the mean square error of sensor predictions by 37% to 50% and KalmanNet improving accuracy by over 60%. In terms of overall model performance, KalmanNet with adaptive feedback achieves an accuracy of 93.2%. Despite its higher computational cost, KalmanNet demonstrates superior precision and robustness under complex state monitoring. The proposed multisensor fusion method not only improves state estimation accuracy and system stability but also offers significant application value for industrial safety monitoring.