Abstract: Aiming at the problem that coal mine dual-wavelength photoelectric smoke sensors suffer zero drift caused by dust accumulation during long-term operation in underground high-dust environments, which seriously degrades the accuracy of fire smoke detection, this paper analyzes the mechanism of zero drift induced by dust accumulation based on light scattering theory with reference to actual underground coal mine working conditions. It is clarified that the damage to the optical characteristics of the reflective bottom surface by dust particles accumulated at the bottom of the optical maze serves as the primary cause of sensor zero drift. Based on the Lorenz-Mie theory, the distribution of dust particles is normalized and simplified, and a unified mathematical model of scattered light intensity affected by deposited dust particles is established. By analyzing the influence of gradual dust accumulation and its distribution variation on independent scattering and secondary scattering, the varying characteristics of scattered light intensity with dust accumulation parameters at different dust deposition stages are summarized. Combined with actual underground coal mine production conditions, the macroscopic variation law of zero drift is induced: the dust deposition process can be divided into three stages, namely the sparse dust accumulation stage, the adjustment stage, and the dense dust accumulation stage. The sensor zero point presents a stepwise linear rise in the sparse dust accumulation stage; in the adjustment stage, the rising rate of the zero point gradually slows down; in the dense stage, the zero point tends to stabilize with slight fluctuating growth, and its specific rising rate is correlated with the height of the dust layer and the distance from the top surface of the dust layer to the photoelectric receiving unit, both of which are affected by the internal dust accumulation rate of the sensor. Industrial tests were carried out with four prototypes arranged at two high-dust measuring points in an underground coal mine. The test results show that the dust accumulation evolution conforms to the variation characteristics of the sparse stage, adjustment stage and dense stage. Based on the above conclusions, ideas for applying the proposed law to zero drift correction and design suggestions for the optical maze of smoke sensors are put forward.