Sprinkling water according to the moisture content is an effective way to suppress dust during coal transportation. However, at present, the moisture content monitoring system of materials in the belt conveyor cannot identify the abnormal points caused by belt vibration. As a result, the predictive model cannot meet the accuracy and range requirements. Therefore, a scheme of moisture content monitoring system in dynamic scenes based on machine learning is proposed. SVM algorithm is used to identify the belt speed and transverse vibration, the 3rd spline interpolation algorithm is used to replace the height and microwave data of the outliers, and then the XGBoost regression model is constructed to predict the moisture content of coal. The results show that, compared with the first generation linear model, the detectable range of coal moisture content is increased from 8.1%~12.3% to 8%~27%, the mean absolute error (MAE) of the model is reduced from 0.86 to 0.53, the mean square error (MSE) is reduced from 0.93 to 0.87, and the model fitting degree (R²) is increased from 0.182 8 to 0.973 5. In the actual monitoring of 75 trains and 14 kinds of coal, the relative error of the predicted value of moisture content is basically within ±5%.