The straightness control of armoured face conveyor is one of the key problems in longwall mining. Aiming at the problems of less sensing information and low accuracy of straightening of armoured face conveyor, a method of straightening based on the movement laws of reverse floating connection mechanism and deep learning trajectory is proposed in VR (Virtual Reality) environment. Firstly, the spatial kinematic model of the reverse floating connection mechanism of roof support and armoured face conveyor are established with knowledge of the industrial robot, the movement laws are obtained. Secondly, SL-LSTM (Single Layer Long Short Term Memory) prediction model is established by sensitivity analysis method after super parameters are selected, and by considering the influence of sensor error and fluctuation of coal floor, the establishment method of date sets and thought of trajectory prediction are put forward. Thirdly, according to requirements for straightness of fully mechanized working face and data availability, the correction model of prediction trajectory and the trajectory-attitude transformation model are established. Finally, based on the obtained armoured face conveyor trajectory, the relative position between a roof support and the corresponding middle trough is determined after timely advancing of roof support, and then the straightening process of armoured face conveyor is realized after precise pushing of the roof support. The straightness error of the straightening method proposed in this paper is within ±0.14 cm and the straightening effect is better by experiment verification.