Aiming at the characteristics of small monitoring range and low alarm accuracy in the traditional fixed monitor of hazardous chemicals warehouse, a patrol robot of hazardous chemicals warehouse is studied. A multi-sensor data fusion method based on laida criterion to improve the fusion performance of BP neural network is adopted. By collecting the data such as the concentration of leaked hazardous chemicals, the ambient temperature and humidity in the warehouse, the data is denoised After normalization, BP neural network is used for fusion output. The prototype test results show that this method can effectively improve the grasp of the space environment of the patrol robot in the dangerous chemical warehouse, greatly improve the accuracy and reliability of the alarm, and have good sensor scalability.