Abstract: Gas solid separation fluidized bed, as an efficient clean coal separation technology, plays an important role in the separation effect due to the transition of flow pattern in the fluidized bed. Rapidly and accurately recognizing the flow pattern of the fluidized bed is of significant importance to enhancing its separation efficiency. Firstly, based on the propagation characteristics of medium wave and voidage wave, a discrimination equation for the transition from fixed bed to bubbling fluidization was established. Research has shown that it was difficult to establish a unified discrimination criterion for flow pattern classification through theoretical calculations; especially under the condition of 2<Re<500, the theoretical calculation method could not achieve effective identification of flow patterns. Secondly, machine learning clustering analysis method was introduced, and seven representative classification algorithms were used. The results showed that the three-layer artificial neural network algorithm exhibited excellent recognition ability for gas-solid fluidized bed flow patterns, with a prediction accuracy of up to 95.8%. Finally, comparing the two flow pattern recognition methods of theoretical calculation and machine learning, the machine learning method not only is easy to operate, but also has higher prediction accuracy, which is conducive to the rapid and accurate identification of fluidized bed flow patterns.