ABSTRACT During the thickening process, the dosage of flocculant has a great influence on the overflow concentration and filter cake moisture. The dosage of flocculant cannot be automatically updated according to the feeding situation, especially when the properties of the concentrated feeding material change, and the overflow concentration detection feedback cannot be adjusted in time. To address these problems, a dynamic concentration detection device was designed. In addition, a dosage of flocculant correction model based on dynamic concentration detection was constructed, and a feed-forwards flocculant additioncontrol model based on the ISSA-LSTM neural network was proposed. The results showed that the relationship between the linear slope of the dynamic concentration and the dosage of flocculant was in line with that obtained using the ExpAssoc model. Then, a double-layer LSTMslime water feedforward flocculant addition control model was proposed based on the multistrategy jointly improved Sparrow SearchAlgorithm (SSA). This model can determine the optimal dosage according to the condition of the infeeding during the thickening process. This new method can improve the intelligence of coal preparation plants and reduce the cost of washing. The scientific application of these research results will help to realize intelligent solid – liquid separation.