Abstract: Abstract: Aiming at the demand for intelligent development of dense medium separation, according to the noise characteristics of dense medium clean coal ash content and the requirements of ash content process control for ash prediction accuracy and prediction time, a time series prediction method of dense medium clean coal ash based on EMD-LSTM is proposed.First, the empirical mode decomposition (EMD) algorithm is used to decompose the different scale components in the clean coal ash time series data step by step to generate a series of eigenmode functions with the same feature scale to remove certain noise effects; secondly, further use Long and short-term memory (LSTM) neural networks can solve the problem of long-term dependence on data, which makes it more prominent in long-term visual field prediction. This method is applied to the short-term prediction of actual data sets. The experimental results show that after optimizing the parameters of the LSTM neural network, the EMD-LSTM-based heavy-medium separation clean coal ash index time series prediction method is obtained by removing the IMF1 component of the model The prediction result has the smallest standard deviation σ (0.1481) and the average absolute error λ (0.1184). The EMD-LSTM model after removing the noise can significantly improve the prediction accuracy and effectively solve the problem of ash prediction of clean coal.